Connect AI agents to the LaunchDarkly MCP server

Check whether a feature flag is ready to be permanently removed from code. Analyzes SDK evaluations to confirm the flag is no longer in use.

Search for LaunchDarkly account members and return their IDs. Supports flexible matching:

Find feature flags that are candidates for cleanup. Returns a prioritized list of stale flags sorted by staleness (worst first). Categories: inactive_30d (no requests in period), launched_no_changes (fully rolled out, no recent changes), never_requested (created but never evaluated). Defaults to showing temporary flags inactive for 30+ days.

Get a specific agent graph by key, including its full edge structure. Each edge connects a source AgentControl Config to a target AgentControl Config with optional handoff data.

Get the latest version of an agent skill by key, including its full SKILL.md markdown, name, description, tags, and version number.

Get detailed configuration for a single AgentControl Config including all its variations. Each variation includes its model, instructions or messages, parameters, attached tools, and judgeConfiguration (attached judges with judgeConfigKey and samplingRate). For judge-mode configs the response also includes evaluationMetricKey.

Health check for an AgentControl Config. Detects common issues: missing models (NO MODEL in UI), missing prompts, orphaned tool references, and empty configs with no variations. Returns a health verdict (healthy, warning, unhealthy) with specific issues and per-variation summaries. Run before updating or experimenting with a config.

Judge scores in one environment, for one AgentControl Config or, without configKey, for every config. Returns one entry per judge metric key ($ld:ai:judge:<suffix>) with `averageScore` and `sampleCount`, covering the built-in judges (accuracy, relevance, toxicity) and every custom judge config in the project. Custom judges include `judgeConfigKey`, `judgeConfigName`, and `isInverted` (true means a lower score is better). Judges with no scores in the window are listed in `noData`. from and to are epoch milliseconds; the default window is the last 7 days. groupBy=variation adds per-variation scores and needs configKey. groupBy=config adds per-config scores. Some apps record judge scores under the judge config's key, not the judged config's key. If a configKey query returns only noData, omit configKey or pass the judge config's key. metricKeys adds judge metric keys that have no judge config, for example from a deleted judge.

Check an AgentControl Config's lifecycle status across all environments. Returns per-environment state (active, inactive, new, launched) and whether targeting is on, plus a summary of which environments the config is active, inactive, or launched in. Useful for understanding where a config is currently being served before making changes.

Read the targeting configuration for an AgentControl Config in a specific environment. Returns variations (with their _id UUIDs and names), individual targets, custom rules, fallthrough (default rule), and off variation. The variation name returned here can be passed as 'variationKey' to update-agentcontrol-config-targeting, update-agentcontrol-config-rollout, and update-agentcontrol-config-targets. Alternatively, use get-agentcontrol-config to obtain the variation's slug key.

Get one AI model config. modelConfigKey is the key from list-agentcontrol-model-configs (for example OpenAI.gpt-4o), not the id. Optional version is an integer from an earlier read (for example 2). Omit it for the latest. Global models accept only version 0 or 1. On a 403, omit version.

Get details about a specific AI Config including its variations and metadata.

Get the health status of an AI Config including latency, error rates, and evaluation metrics.

Get the status of an AI Config across all environments in a project.

Get the targeting rules and rollout configuration for an AI Config in a specific environment.

Get a single AI tool definition including its full schema. Use to inspect a tool's parameters before attaching it to an AgentControl Config variation.

Get detailed information about a specific observability alert.

Show when an alert actually fired, and what state it is in now. This is the tool that answers "is this alert doing anything?". list-alerts cannot: disabled=false only means the alert is being evaluated, and updated_at is the last config edit, not the last firing. Use this to find stale alerts worth retiring, to check whether a threshold is reachable, and to confirm an alert still works after you retune it. INPUTS: - projectKey (required): The LaunchDarkly project key (e.g., "default"). - alertId (required): The numeric ID of the alert (from list-alerts) - lookbackDays (optional, default 30): How far back to count firings. Must be positive. - limit (optional, default 20, max 100): Max individual firing rows to return, newest first RETURNS: - firing_count: Total firings in the window. May exceed the rows returned — check truncated - last_fired_at: Timestamp of the most recent firing, or null if it never fired in the window - firings: Newest-first rows with timestamp, state, group_by_key, and investigation_conversation_id when an automatic investigation ran for that firing - current_state: Latest state per group (any state, unwindowed). This is how you tell an alert firing RIGHT NOW from one that fired and resolved. States: Normal, Pending, Alerting, AlertingSilently, NoData, Error Only Alerting transitions are counted, so firing_count is a count of real firings rather than of every state change. Retention is not a constraint: the backing table has no TTL, so 30-, 60-, and 90-day windows are all equally valid. Pick the one that fits the question. A current_state of NoData is a finding worth reporting — the alert's query matches nothing, so it can no longer fire regardless of threshold. That usually means the underlying service was renamed or stopped emitting. Example usage: {"projectKey": "default", "alertId": 42, "lookbackDays": 90}

Get all stored instances of a specific context by kind and key. Returns every recorded occurrence of the context, including its full attribute set (e.g., merchantCountryCode, disbursementRail) and which SDK/application reported each instance.

Fetch the full markdown content of a LaunchDarkly documentation page.

Deprecated: use get-release-autopilot-repo-settings instead. Get Release Autopilot settings for one GitHub repository (effective auto-flagging / auto-releasing). Pass repo as owner/name, or githubRepoId. 404 means the repo is not mapped yet.

Deprecated: use get-release-autopilot-settings instead. Get account-wide Release Autopilot settings: whether auto-flagging and auto-releasing are on, and whether auto-flagging pull requests need approval. These are the master gate — a mapped repository cannot enable a capability the account has off. Requires LD-API-Version beta (sent automatically). Do not use autoCleanup; it is not part of Release Autopilot settings yet.

Get full details about a specific feature flag including all variations, targeting rules across environments, and metadata.

Get health indicators for a feature flag including evaluation counts, error rates, and usage patterns.

Get the on/off status and targeting summary for a feature flag across all environments in a project.

Get unique-context exposure counts (distinct-context cardinality) for a flag in an environment. Returns how many distinct users/accounts/etc. were actually exposed to the flag over a time window, broken down by context kind. CRITICAL DISTINCTION: unique exposures are NOT evaluation counts. One context evaluated 10,000 times counts as 1 exposure (not 10,000). This ~1000x difference is critical for guarded rollout sizing. The guarded-release UI gates on this figure. For total SDK call counts use get-flag-evaluation-summaries. from/to accept ISO 8601 strings or Unix millisecond timestamp strings (default: last 30 days).

Get the profile of the currently authenticated member, including their role and permissions.

Get details about a specific LLM playground, including its variants (evaluation definitions and their positions).

Get a specific prompt snippet by key. Returns the snippet's name, text content, tags, version, and creation time.

Get Release Autopilot settings for one GitHub repository (effective auto-flagging / auto-releasing). Pass repo as owner/name, or githubRepoId. 404 means the repo is not mapped yet.

Get account-wide Release Autopilot settings: whether auto-flagging and auto-releasing are on, and whether auto-flagging pull requests need approval. These are the master gate — a mapped repository cannot enable a capability the account has off. Requires LD-API-Version beta (sent automatically). Do not use autoCleanup; it is not part of Release Autopilot settings yet.

List the views a flag, segment, or AgentControl Config is linked to. Views are an Enterprise feature. Flags and AgentControl Configs are identified by key; segments additionally require `environmentId` because segments are environment-scoped.

Get detailed configuration for a single segment in an environment. Returns rules, included targets, excluded targets, tags, and creation date. Use to verify segment state after creation or updates.

Get a time series of service connection minutes from your LaunchDarkly account. Use this for capacity planning, cost analysis, and before/after deploy comparisons (e.g. compare daily incremental minutes around a deploy date).

Get a signed download URL for one rrweb event chunk of a session. The URL is valid for ~15 minutes and points to Brotli-compressed JSON — an array of raw rrweb events (DOM snapshots, mutations, interactions). Decompress with Brotli before parsing. REQUIRED INPUTS: - projectKey: The LaunchDarkly project key (e.g., "default"). - sessionSecureId: The secure ID of the session. REQUIRED. - index: The chunk_index of the chunk to download (from get-session-event-chunks). REQUIRED. Chunk payloads can be large; prefer query-timeline-events for a cheap chronological overview and only fetch raw chunks when you need actual DOM/interaction detail. Here's an example of a valid input: {"projectKey": "default", "sessionSecureId": "abc123def456", "index": 0}

List the rrweb event chunks recorded for a session. Session replay events are stored in chunks; this returns one entry per chunk with chunk_index and the chunk's start timestamp (epoch millis). Use it to see how a session's recording is segmented, then fetch a specific chunk's raw contents with get-session-event-chunk-url. REQUIRED INPUTS: - projectKey: The LaunchDarkly project key (e.g., "default"). - sessionSecureId: The secure ID of the session to list event chunks for. REQUIRED. The session's chunked field (from query-sessions) tells you whether chunks exist — unchunked (very short or old) sessions return an empty list. For a cheap chronological overview of a session prefer query-timeline-events; only fetch raw chunks when you need actual DOM/interaction detail. Here's an example of a valid input: {"projectKey": "default", "sessionSecureId": "abc123def456"}

Get a signed download URL for the rrweb event chunk covering a specific timestamp within a session. Picks the chunk whose start is at or before the timestamp — useful to jump straight to the moment an error or interaction happened without listing all chunks first. The URL is valid for ~15 minutes and points to Brotli-compressed rrweb events JSON. REQUIRED INPUTS: - projectKey: The LaunchDarkly project key (e.g., "default"). - sessionSecureId: The secure ID of the session. REQUIRED. - timestamp: Timestamp of interest in ISO format (e.g., "2024-01-15T10:00:00Z"). REQUIRED. Here's an example of a valid input: {"projectKey": "default", "sessionSecureId": "abc123def456", "timestamp": "2024-01-15T10:00:00Z"}

Discover the observed values for ONE attribute key, ranked by frequency. The companion to get-keys: that tells you a key exists, this tells you what to compare it against. Use it whenever a key=value filter returns 0 rows but you believe the key is right; the value vocabulary is usually the problem. Two things this is especially good for: - Service names. service_name values often carry environment prefixes (production-, staging-) but not uniformly. Don't guess the prefix; enumerate. - Confirming a guess is wrong. If a search for a service found nothing, {"key": "service_name", "search": "checkout"} shows what actually exists instead of widening the time range pointlessly. A long value list means the key is high-cardinality (e.g. trace_id) and is not useful to filter or group on; pick a coarser key. INPUTS: - projectKey (required): The LaunchDarkly project key (e.g., "default"). - productType (required): One of "logs", "traces", "sessions", "errors", "metrics", "paevents", "events". Use "paevents" for product-analytics events; "events" is the legacy session_events source. - key (required): The attribute key whose values to list (e.g., "service_name") - startDate (required): ISO format start date (e.g., "2024-01-15T10:00:00Z") - endDate (optional): ISO format end date (defaults to current time) - search (optional): Substring filter over values. Substring, not fuzzy; search a short stem - limit (optional, default 30, max 100): Maximum number of values to return RETURNS: - values: Array of { value, count }, most frequent first - product_type, key, value_count - _hint: Present when no values were found Example: {"projectKey": "default", "productType": "traces", "key": "service_name", "startDate": "2024-01-15T10:00:00Z"}

Get a single view by key. Views are an Enterprise feature. Returns name, description, tags, maintainer, and linked-resource counts (flags, segments, AgentControl Configs). Use list-view-resources to enumerate the linked resources themselves.

Get detailed information about a specific dashboard including all its graphs.

Get details about a specific offline dataset by ID, including its processing status and row count.

Get details about a specific environment in a project, including its keys and settings.

Get details about a specific evaluation definition, including its configuration and last run status.

Get the summary results of a completed evaluation run, including pass/fail counts and aggregate scores.

Get details about a specific experiment including its configuration, current iteration status, and metric assignments.

Get detailed statistical results for a single metric on an experiment iteration. Returns, per treatment: sample sizes (analyzedUnitCount, trafficCount, conversionCount), the observed mean and standard deviation, and a `statistics` block with the lift versus control (relativeDifferences: estimate plus lower/upper interval bounds and, for frequentist experiments, a per-comparison pValue), the overall pValue, Bayesian probabilities (probabilityToBeBest, probabilityToBeatBaseline, expectedLoss), and an isSignificant flag. Reads from LaunchDarkly's internal analysis API. Defaults to the current iteration and the latest analysis; pass iterationId to target a previous iteration. Use disableMultipleComparisonCorrection / disableSequentialTesting to see uncorrected numbers. Use get-experiment-results for a cross-metric overview, or get-experiment to discover metric keys.

Get an overall results summary for an experiment, across every metric on the current iteration. Discovers the iteration's metrics, then fetches each metric's analysis from LaunchDarkly's internal results API and returns a compact per-metric summary: which treatment is leading (isBest) — the arm with the highest probability to beat the baseline for Bayesian experiments, or the lowest p-value versus control for frequentist (note: not significance-gated, so a treatment can be flagged leading even when nothing is significant) — each treatment's lift versus control, whether it's statistically significant, and sample sizes. Pass iterationId to summarize a previous iteration. For the full per-treatment statistics (intervals, p-values, Bayesian probabilities) on one metric, use get-experiment-metric-results.

Discover available data keys/dimensions for a product type.

Get details about a specific metric including its configuration and associated experiments.

Get details about a specific LaunchDarkly project, including its environments and settings.

Check whether any SDKs have been active in a given environment. Returns true if any SDK has initialized or sent events in the environment. Use this to verify that an environment is actually in use before cleanup. Optionally filter by sdkName or sdkWrapperName to check a specific SDK.

List all agent graph definitions in a project.

List the AgentControl Config variations that attach an agent skill, with the skill version each one is pinned to. Use before updating or deleting a skill.

List the versions of an agent skill, newest first. Returns each version's number, name, description, tags, and creation time.

List agent skills in a project. Agent skills are versioned SKILL.md documents that AgentControl Config variations attach by {key, version}. Returns key, name, description, latest version, and tags; use get-agent-skill to read a skill's markdown. Use query to search by key, name, or description.

List approval requests for an AgentControl Config. Shows status, review state, and who requested each change. Omit env (or pass '*') to list project-scoped variation approval requests; pass a specific env key to filter to targeting-change requests for that environment. Use to check on approval progress after creating a request.

Search and browse AgentControl Configs in a project. Returns a paginated list with key, name, mode (agent, completion, or judge), tags, variation count, and a `usedInGraphs` array listing the agent graph keys that reference each config (as the graph root or as an edge source/target). An empty `usedInGraphs` means the config is not referenced by any graph (orphaned). If graph usage cannot be loaded, `usedInGraphs` is omitted entirely (rather than empty) so an absent field is never mistaken for orphaned. Use query to search by name or key, tags to filter by tag, or view to only include configs linked to a view (Views are an Enterprise feature).

List the AI model configs in a project: global catalog models (global=true) and custom models (global=false). Pass an item's key (for example OpenAI.gpt-4o) as modelConfigKey and its id (for example gpt-4o) as modelName to create-agentcontrol-config-variation. Optional restricted: true returns only restricted models, false returns only unrestricted models. Omit it for all. Costs are USD per token.

List all AI Configs in a project with their current status and variation count.

List the versions of one AI tool in a project, newest first. Returns each version's key, description, schema, version, tags, and customParameters. Use to pick a specific version to pass in the tools array ({key, version}) with create-agentcontrol-config-variation or update-agentcontrol-config-variation. Pagination: responses include `totalCount`. To get the next page, set `offset` to the current offset + limit.

List AI tool definitions in a project. Returns each tool's key, description, and schema. Tools are attached to AgentControl Config variations to give models function-calling capabilities.

List existing observability alerts for the project.

List pending approval requests for a flag in an environment. Shows status, review state, and who requested the change. Use to check on approval progress after creating a request.

Deprecated: use list-release-autopilot-github-repos instead. List GitHub repositories on this account's LaunchDarkly GitHub App install, with display name (usually owner/name) and numeric githubRepoId. Use this before mapping a repo. This is not the list of already-mapped repos (that is list-release-autopilot-repo-settings). Optional projectKey: any project works; the install is account-wide (same as the settings UI). When githubConnectUrl is present, give that link to the user and retry after they finish. Requires Observability GraphQL (O11Y_GRAPHQL_ENDPOINT) like other Vega GitHub tools.

Deprecated: use list-release-autopilot-repo-settings instead. List Release Autopilot settings for GitHub repositories already mapped to a LaunchDarkly project. Unmapped install repos are omitted — use list-release-autopilot-github-repos for those. Each automation is the effective value (account + repo override). githubRepoId is numeric.

List LLM playgrounds in a project. Returns playground names, variant counts, and timestamps. Supports search by name and pagination.

List LaunchDarkly projects in the account. Use to discover project keys before calling project-scoped tools. Filtering: use `query` to search by project name or key (case-insensitive), `tags` to filter by tag (all must match). Sorting: use `sort` with `name` or `-name`. Pagination: responses include `totalCount` and `pageInfo.offset`. To get the next page, set `offset` to the current offset + limit.

List prompt snippets in a project. Prompt snippets are reusable text blocks that can be referenced inside AgentControl Config variation prompts to keep common instructions consistent. Returns key, name, text, version, and tags.

List GitHub repositories on this account's LaunchDarkly GitHub App install, with display name (usually owner/name) and numeric githubRepoId. Use this before mapping a repo. This is not the list of already-mapped repos (that is list-release-autopilot-repo-settings). Optional projectKey: any project works; the install is account-wide (same as the settings UI). When githubConnectUrl is present, give that link to the user and retry after they finish. Requires Observability GraphQL (O11Y_GRAPHQL_ENDPOINT) like other Vega GitHub tools.

List Release Autopilot settings for GitHub repositories already mapped to a LaunchDarkly project. Unmapped install repos are omitted — use list-release-autopilot-github-repos for those. Each automation is the effective value (account + repo override). githubRepoId is numeric.

List segments in a project environment. Returns a paginated list with rule counts and target counts. Use query to search by name or key, tags to filter by tag, or view to only include segments linked to a view (Views are an Enterprise feature). Always list first to avoid creating duplicates.

List the resources of one type linked to a view. Views are an Enterprise feature. Returns resource keys with link timestamps (and environment info for segments). Use query for a case-insensitive search on resource keys. Sorting: use `sort` with `linkedAt` (default) or `name`. Pagination: responses include `totalCount` and `pageInfo.offset`. To get the next page, set `offset` to the current offset + limit.

List views in a project. Views are an Enterprise feature. Returns a paginated list with key, name, description, tags, maintainer, and (when includeResourceSummary is set) linked-resource counts. Use query to search by name, tags to filter by tag. Sorting: use `sort` with `key`, `name`, or `updatedAt` (default `updatedAt`, most recent first). Pagination: responses include `totalCount` and `pageInfo.offset`. To get the next page, set `offset` to the current offset + limit.

List existing dashboards (visualizations) for the project.

List offline datasets for a project. Returns id, name, status, row count, and creation info for each dataset.

List AI Config evaluations in a project. Returns evaluation definitions with their names, associated config keys, and last run info.

List experiments in a project, optionally filtered by environment. Returns key, name, description, and current iteration status for each experiment.

Search and browse feature flags in a project. Returns a paginated list scoped to a single environment.

List recent event keys received by a LaunchDarkly project — used to check which events are actively flowing before creating a metric. Returns up to 50 event keys with last-seen timestamps. If the event key you need isn't here, it may not be instrumented yet.

List metrics in a LaunchDarkly project. Returns key, name, measureType (count/occurrence/value), eventKey, successCriteria, tags, and how many flags each metric is attached to.

List release policies for a project. Each policy defines preferred release methods and the metrics or metric groups that automatically attach to guarded rollouts when the policy's conditions match (e.g. specific environments or flag tags).

Resolve which release policies would govern a flag in a given environment — a read-only dry-run of the server-side policy matching logic. Does not create or modify anything.

Preview a chart/graph inline WITHOUT saving it to a dashboard. Always use this tool first when a user asks to create or visualize a chart. Returns both the graph configuration and the queried metrics data so the user can see the chart rendered inline. After showing the preview, ask the user if they want to save it to a dashboard (using create-graph).

Query the audit log to find what changed in a LaunchDarkly account.

Query the service dependency map (service map) for a project. Returns the directed call graph between services, derived from trace data over the given time window. REQUIRED INPUTS: - projectKey: The LaunchDarkly project key (e.g., "default"). - startDate: Start date in ISO format (e.g., "2024-01-15T10:00:00Z"). REQUIRED. OPTIONAL INPUTS: - endDate: End date in ISO format (e.g., "2024-01-15T11:00:00Z"). If not provided, uses current time. DATE FORMAT: - Use ISO format: "2024-01-15T10:00:00Z" or "2024-01-15T10:00:00-08:00" - Dates should be reasonable and within your data retention period - Always specify timezone if not using Z (UTC) RETURNS: - services: Sorted list of every service name (graph nodes), including standalone services with no observed dependencies. - dependencies: Directed edges as { source, target } pairs, where `source` calls `target`. - service_count / dependency_count: Convenience counts. Examples: 1. Specific time window: {"projectKey": "default", "startDate": "2024-01-15T10:00:00Z", "endDate": "2024-01-15T11:00:00Z"} 2. From specific time to now: {"projectKey": "default", "startDate": "2024-01-15T10:00:00Z"}

Retrieve bucketed, aggregated values over time for a product type. Returns time-series buckets (not raw events) suitable for charting trends, comparing groups, and computing sums/averages/percentiles. Use this instead of query-logs/query-traces/query-sessions/query-error-groups when you want aggregated values over a window rather than individual records.

Query project error groups. Retrieve error groups for a given project with explicit date range parameters.

Query flag evaluations for a session. Retrieve flag evaluation events for a specific session to understand which feature flags were evaluated during that session.

Query project logs. Retrieve logs for a given project with explicit date range parameters.

Query project sessions. Retrieve sessions for a given project with explicit date range parameters.

Query timeline indicator events for a session. Retrieve timeline indicator events for a specific session to understand what happened during that session. Keep in mind that this will not include flag evaluation events, you must use the query-flag-evaluations tool to retrieve those.

Query project traces. Retrieve traces for a given project with explicit date range parameters.

Search for stored LaunchDarkly contexts in an environment by kind, key, or attribute value. Returns context records including their stored attributes (e.g., merchantCountryCode, disbursementRail) and when the context was last seen.

Search LaunchDarkly documentation to retrieve authoritative reference material for flags, targeting rules, experiments, segments, AgentControl configs, SDKs, API fields, rollouts, metrics, and more.

Apply an already-approved AgentControl Config approval request. This executes the changes that were approved by reviewers. Only works on requests with reviewStatus 'approved'. Does NOT approve requests: that must be done by a human reviewer. Equivalent to apply-approval-request but scoped semantically to AgentControl Configs.

Check whether a feature flag is fully rolled out to a single deterministic variation across all traffic-bearing environments. This detects the important stale case that inactive detection misses: a flag at 100% rollout is still evaluated on every request (never looks inactive), but its code branch is dead and it is prime for removal. ALGORITHM: For each environment, checks if serving is deterministic (targeting off → offVariation, or targeting on with no rules/targets/prerequisites/rollout → simple fallthrough variation). Traffic-bearing environments are those with a recent lastRequested timestamp. A flag is fully rolled out when all traffic-bearing envs converge on the same variation AND a stability gate is met (at least one env is 'launched' or the flag is older than stabilityDays). Returns isFullyRolledOut, consensusVariation (value + name when fully rolled out), per-environment results, and blockers explaining any failure to be fully rolled out.

Clone an existing AgentControl Config variation with selective overrides. Reads the source variation, applies any provided overrides (model, instructions, messages, parameters, tools), and creates a new variation. Returns both the source and created variation so you can compare the diff. Use this for A/B experimentation: change one thing at a time while keeping everything else constant.

Clone an existing AI Config variation to create a new variation with the same settings.

Create a new agent graph in a project. An agent graph defines a directed graph of AgentControl Configs for multi-agent workflows. Provide a rootConfigKey and edges to define the graph structure, or create the graph with just metadata and add edges later via update-agent-graph. If rootConfigKey or edges are provided, both must be present.

Create an agent skill from SKILL.md content. Provide a unique key, a display name, and the verbatim markdown. The skill starts at version 1. Attach it to a variation with create-agentcontrol-config-variation or update-agentcontrol-config-variation by passing skills: [{key, version}].

Create a new AgentControl Config in a project. This creates the config shell: use create-agentcontrol-config-variation next to add a model, prompts, and parameters. Mode determines whether variations use 'instructions' (agent), 'messages' (completion), or 'messages' (judge). Judge-mode configs evaluate other AgentControl Config outputs. IMPORTANT: evaluationMetricKey (format: $ld:ai:judge:<suffix>) is REQUIRED when mode is 'judge'; omitting it will cause a 400 error. For agent and completion mode it is not accepted. Pass viewKeys to link the new config to one or more views at creation time (Views are an Enterprise feature).

Create an approval request for an AgentControl Config change. Use this for variation-level edits (model, prompts, parameters, tools via instructions like updateAIConfigVariation, createAIConfigVariation, deleteAIConfigVariation) as well as targeting changes (see update-agentcontrol-config-targeting, update-agentcontrol-config-rollout, update-agentcontrol-config-targets). Omit env (or pass '*') for variation-level changes — they are project-scoped, not env-scoped. Pass a specific env key for targeting changes. The request is reviewed by approvers before taking effect. Requires at least one reviewer: notifyMemberIds and/or notifyTeamKeys. If the user named a person, resolve them with find-members and pass the returned id. If they did not name anyone, ask who should review before calling this tool. Does NOT approve the request: a human reviewer must do that. Once approved, use apply-approval-request to execute the change.

Create a variation for an AgentControl Config. A variation defines the model, prompts, parameters, and tools. modelConfigKey must be in Provider.model-id format (e.g. OpenAI.gpt-4o, Anthropic.claude-sonnet-4-5) for models to display correctly in the UI. Agent-mode configs use 'instructions' (a string); completion-mode configs use 'messages' (an array of {role, content} objects). To attach AI tools, pass the 'tools' array with {key, version} entries (create tools first with create-ai-tool). To attach judges, pass judgeConfiguration with a judges array of {judgeConfigKey, samplingRate} entries.

Create a custom AI model config for a model that is not in the global catalog. Call list-agentcontrol-model-configs first, and use a global model if one exists. Required: key, name, id. key uses letters, numbers, ., _, and - (for example gpt-image-2-20260916-uzw1la), not a global key. id is the provider's model ID (for example gpt-image-2). provider matches the key's prefix, if any (for example OpenAI). Costs are USD per token (for example 0.000005). params and customParams are freeform JSON objects (for example {"temperature": 0.7}). Set maintainerId or maintainerTeamKey, not both. Next, pass key as modelConfigKey and id as modelName to create-agentcontrol-config-variation.

Create a new AI Config in a project. An AI Config manages AI model configurations with feature flag-style targeting and experimentation.

Create a new variation for an AI Config with specific model, prompt, and parameter settings.

Create a new AI tool definition in a project. The schema should be a raw JSON Schema object with type, properties, and required fields (e.g. {"type": "object", "properties": {...}}). Do NOT use the OpenAI function calling wrapper format. After creation, attach the tool to a variation using update-agentcontrol-config-variation.

Create a new observability alert that fires when a metric crosses a threshold.

Create an approval request for a flag change in an environment that requires approvals. Provide the same semantic patch instructions you would use for a direct change. The request will be reviewed by approvers before taking effect. Does NOT approve the request: that must be done by a human reviewer.

Record an automated rollout config in LaunchDarkly for a feature-flagged change.

Create a new feature flag in a project. Defaults to a boolean temporary flag. After creation the flag is OFF in all environments: use toggle-flag to enable it.

Create a new LLM playground. A playground lets you compare AgentControl Config variations side-by-side. Provide a name and variants — each variant references an evaluation definition (by evaluationId) and a display position.

Create a new reusable prompt snippet.

Create a new segment in a project environment. The segment is empty after creation — use update-segment-rules to add targeting rules or update-segment-targets to add individual context keys. Segment keys are immutable after creation: choose carefully. Pass viewKeys to link the new segment to one or more views at creation time (Views are an Enterprise feature).

Create a new view in a project. Views are an Enterprise feature. The view is empty after creation — use link-view-resources to link flags, segments, or AgentControl Configs to it, or pass viewKeys when creating those resources. The key is converted to lowercase by the API, so use the lowercase form to reference the view afterward. MAINTAINER: set maintainerTeamKey or maintainerId to assign an owner; only one of the two can be specified. Prefer maintainerTeamKey when a suitable team exists.

Create a new empty dashboard (visualization) for organizing charts.

Create a new offline dataset for AI evaluation. Provide a dataset name, filename, and format (csv, json, or jsonl). The dataset is created in pending status and must be uploaded separately. Returns the dataset ID and upload URL.

Create a new AI evaluation definition. An evaluation defines a comparison between AI Config variations using a dataset and judge criteria. After creation, use run-evaluation to start an evaluation run.

Create a new experiment on a flag or AgentControl Config. An experiment measures the impact of different variations on specified metrics. You must provide the initial iteration definition including a hypothesis, metrics, treatments, and flag configuration. One treatment must be marked as the baseline. Use `methodology` to pick the results-analysis approach (`bayesian` default, `frequentist`, or `export_only` for data-export-only mode). Use `analysisConfig` to set thresholds and multiple-comparison correction. Use `iteration.attributes` to declare slicing dimensions and `iteration.covariateId` to enable stratified sampling.

Add a chart/graph to an existing dashboard.

Create a new metric in a LaunchDarkly project.

Create a new LaunchDarkly project. Projects are top-level containers for feature flags and environments.

Invite one or more people to the LaunchDarkly account by email. Each invitee receives an email invitation to join. Optionally assign a role (reader, writer, admin) — defaults to reader if not specified.

Link flags, segments, or AgentControl Configs to a view. Views are an Enterprise feature. IDENTIFYING RESOURCES: - flags / aiConfigs: pass `keys` - segments: pass `segmentIdentifiers` ({environmentId, segmentKey} pairs — segments are environment-scoped) - any type: pass `filter` to link everything matching a list-endpoint filter expression (flags use comma-separated field:value filters, segments use queryfilter syntax; `environmentId` is required with a segments filter). Keys and filter combine as a union. Large key lists are batched automatically at 20 keys per API request. The response aggregates successCount/failureCount across batches and lists any resources that failed to link.

Restrict AI model configs so that variations cannot select them. keys are model config keys from list-agentcontrol-model-configs (for example ["OpenAI.gpt-4o"]), not ids. Returns the restricted keys and per-key errors. For 'key is already in use', move the variations to another model with update-agentcontrol-config-variation, then retry. For 'unknown model config key', check list-agentcontrol-model-configs. Undo with unrestrict-agentcontrol-model-configs.

Start a new run of an AI evaluation. This executes the evaluation against the configured dataset and judge criteria. Returns the run ID and status.

Create an AgentControl Config with its first variation in one step. This is the recommended way to set up a new AgentControl Config: it creates the config, adds a variation with model and prompts, and verifies everything is configured correctly. Returns the full config detail with variation. To attach AI tools, pass the 'tools' array with {key, version} entries. IMPORTANT: evaluationMetricKey (format: $ld:ai:judge:<suffix>) is REQUIRED when mode is 'judge'. Pass viewKeys to link the new config to one or more views at creation time (Views are an Enterprise feature). Use create-agentcontrol-config + create-agentcontrol-config-variation separately only when you need more control.

Initialize an AI Config with its first variation and targeting setup in one step.

• Idempotent

Start a guarded rollout on a flag's default rule (fallthrough). A guarded rollout progressively increases traffic to the test variation through a series of stages while monitoring metrics for regressions. Each stage specifies a rolloutWeight (percentage in thousandths, e.g. 10000 = 10%) and a monitoring window in milliseconds. Metrics define what to monitor and whether to notify or rollback on regression. Requires the flag to be on and not already have an active guarded rollout. If the environment requires approval, the response includes requiresApproval: true and the attempted instructions; call `create-approval-request` with the returned instructions to submit it for review.

Turn an AgentControl Config's targeting on or off in a specific environment. Returns the previous and new state. This is equivalent to 'turnFlagOn' / 'turnFlagOff' for feature flags. If the environment requires approval, the response includes requiresApproval: true and the attempted instructions; call `create-agentcontrol-config-approval` with the returned instructions to submit it for review.

• Idempotent

Enable or disable an observability alert without changing its configuration.

• Idempotent

Turn a feature flag on or off in a specific environment. When off, all users receive the off variation.

• Idempotent

Remove AI model configs from the restricted list so that variations can select them again. keys are model config keys (for example ["OpenAI.gpt-4o"]), not ids. To see restricted keys, call list-agentcontrol-model-configs with restricted=true.

Update an agent skill's markdown, name, description, tags, or maintainer. Only provided fields are updated. Variations stay pinned to the version they attached; use list-agent-skill-references to find them and update-agentcontrol-config-variation to move them to the new version.

Update an AgentControl Config's metadata: name, description, tags, or archive status. Does NOT modify variations: use update-agentcontrol-config-variation for model, prompt, or parameter changes. Set archived: true to archive (reversible).

• Idempotent

Update the default (fallthrough) rule for an AgentControl Config in an environment. Set a percentage rollout across variations, or serve a single variation to all unmatched contexts. Weights must sum to 100. Use human-friendly percentages (e.g., 80 for 80%). Accepts a variation key (from get-agentcontrol-config) or variation name (from get-agentcontrol-config-targeting); the tool resolves either to the internal variation _id (UUID) before patching. If the environment requires approval, the response includes requiresApproval: true and the attempted instructions; call `create-agentcontrol-config-approval` with the returned instructions to submit it for review.

• Idempotent

Add, remove, or modify custom targeting rules for an AgentControl Config in an environment. Rules are evaluated top-to-bottom; first matching rule wins. Supported kinds: addRule, removeRule, addClauses, removeClauses, reorderRules, replaceRules, updateRuleDescription, addValuesToClause, removeValuesFromClause, updateClause. Use get-agentcontrol-config-targeting first to obtain ruleId, clauseId, and variationId values referenced by these instructions. If the environment requires approval, the response includes requiresApproval: true and the attempted instructions; call `create-agentcontrol-config-approval` with the returned instructions to submit it for review.

Add or remove specific contexts from individual AgentControl Config targeting in an environment. Individual targets are the highest priority and override all rules. Supported kinds: addTargets, removeTargets, clearTargets, replaceTargets. Each instruction takes a 'variationKey' set to either the variation key (from get-agentcontrol-config) or the variation name (from get-agentcontrol-config-targeting); the tool resolves either to the internal variation _id before patching. If the environment requires approval, the response includes requiresApproval: true and the attempted instructions; call `create-agentcontrol-config-approval` with the returned instructions to submit it for review.

Update a custom AI model config. modelConfigKey is the key from list-agentcontrol-model-configs (for example OpenAI.gpt-image-2), not the id. Send at least one field. Only sent fields change, and params, customParams, and tags replace the current values. Formats match create-agentcontrol-model-config. provider and id cannot change: call create-agentcontrol-model-config with a new key instead. Global models (global=true) cannot be updated.