LaunchDarkly MCP (official)
Capped at 70: no passing tool call yet; the tests add up to 100. Why scores are capped
Test points add up to 100; the score is capped at 70 until a tool call passes. Not a quality or security rating. Methodology
Test points before the cap (100/100)
- Install20/20
- Handshake30/30
- Maintenance25/25
- Docs15/15
- License10/10
Tested npm @launchdarkly/mcp-server 0.7.2 (MIT; repo launchdarkly/mcp-server). `mcp start` over stdio listed 30 tools for feature flags, AI configs, metrics and experiments without a key. Both read calls (list-feature-flags, get-environments) answered 'Invalid account ID header': every tool needs a LaunchDarkly API access token. Core tools not tested. Create, update and delete tools were not called.
Tool-call test (2026-11-11, direct MCP stdio client, no LLM): fail. 0 of 2 calls passed; every tool needs a LaunchDarkly API key. Core tools not tested. Evidence: local test run, Nov 11, 2026.
Transport (tested 2026-11-11): stdio
Token cost (estimate): approx. 6634 tokens, cl100k estimate (30 tools, measured 2026-11-11, version 0.7.2). This is an estimate from our testing of the tool names, descriptions and input schemas, counted with cl100k_base.
Permissions: 15 read, 15 write (state-changing), 0 unclassified · Auth: API key · Rate limit: unknown · Pricing unit: unknown
Auth: API key. Rate limit: unknown. Pricing unit: unknown.
create-ai-config— write (state-changing)create-ai-config-variation— write (state-changing)create-experiment— write (state-changing)create-experiment-iteration— write (state-changing)create-feature-flag— write (state-changing)create-metric— write (state-changing)delete-ai-config— write (state-changing)delete-ai-config-variation— write (state-changing)delete-feature-flag— write (state-changing)get-ai-config— readget-ai-config-targeting— readget-ai-config-variation— readget-audit-log-entries— readget-code-references— readget-environments— readget-experiment— readget-feature-flag— readget-flag-status-across-environments— readget-metric— readlist-ai-configs— readlist-experiments— readlist-experiments-any-environment— readlist-feature-flags— readlist-metrics— readupdate-ai-config— write (state-changing)update-ai-config-targeting— write (state-changing)update-ai-config-variation— write (state-changing)update-experiment— write (state-changing)update-feature-flag— write (state-changing)update-metric— write (state-changing)
listing-flags calls_need_key or auth_kind
Score 70/100 · Tools 30 · Token cost approx. 6,634 tokens · Compare · Last tested 2026-11-11 · Version tested 0.7.2 · Package @launchdarkly/mcp-server · Repo launchdarkly/mcp-server
Summary: LaunchDarkly MCP (official) completed an MCP handshake without credentials and exposed 30 tools.
Evidence summary
- Handshake
- pass
- Tools listed
- 30
- Context cost
- approx. 6634 tokens, cl100k estimate (30 tools, measured 2026-11-11, version 0.7.2; an estimate) ranking · add up your servers' context cost
- API key
- Connects without a key; tool calls need one
- Tested
- 2026-11-11
Test results
| Test | Result | Points | Evidence |
|---|---|---|---|
| install | pass | 20 | npm install @launchdarkly/mcp-server (Node 20, Linux) succeeded in 1s. |
| handshake | pass | 30 | MCP initialize + tools/list over stdio succeeded with no credentials; 30 tools listed in 6.3s. |
| maintenance | pass | 25 | Repo pushed 0 days before test; 27 GitHub stars. |
| docs | pass | 15 | README over 3,000 chars; install instructions present; mcpServers config example present. Check of the package's own README (README shipped in npm package). |
| license | pass | 10 | MIT (OSI-approved). |
Local-model readiness
Token cost (estimate): approx. 6634 tokens, cl100k estimate for 30 tools · about 81% of an 8k context. This is an estimate from our testing (counted with cl100k_base).
Small-model friendly (schema heuristic, not a model result): unknown.
Not tested
Small-model task runs: Not yet tested.
What we tested and what we didn't
Tested on 2026-11-11
- Install from the registry
- We connected with no credentials and asked for its tool list: it listed 30 tools (Method details: MCP
initializethentools/list) - Maintenance signals (last push, archived/deprecated status)
- Docs (README length, install and config example)
- License (OSI-approved or not)
- Token cost (estimate)
Not tested
- Security: not audited (no code review, no permission or data-flow analysis)
- Real tasks (no functional task runs yet)
- Behaviour with real credentials or a real account
- Performance, rate limits or load
- Correctness of results beyond the listed tasks
Methodology: how we test. A high score means it installs and responds, not that giving it access to your data carries no risk.
Known security issues
Checked on 2026-11-11 for version 0.7.2 against OSV, GitHub advisories, GitHub repository advisories. Scope: the package itself; dependencies not scanned.
No published advisories found in OSV and GitHub advisories for @launchdarkly/mcp-server 0.7.2 as of 2026-11-11. This is not a security audit.
Monthly roundup: MCP security issues, October 2026.
Tool list changes
Baseline recorded from the 2026-11-11 test run; changes appear after the next test.
Snapshot hash (fingerprint of the tool list)
b512e2e14c68: SHA-256 of the canonical tool names, descriptions and input schemas. If it changes, the tool list changed.
Raw test record (JSON)
{
"name": "@launchdarkly/mcp-server",
"version": "0.7.2",
"registry": "npm",
"status": "working",
"tested": "2026-11-11",
"tools_count": 30,
"deprecated": false,
"archived": false,
"tests": {
"install": [
"pass",
20,
"`npm install @launchdarkly/mcp-server` (Node 20, Linux) succeeded in 1s."
],
"handshake": [
"pass",
30,
"MCP `initialize` + `tools/list` over stdio succeeded with no credentials; 30 tools listed in 6.3s."
],
"maintenance": [
"pass",
25,
"Repo pushed 0 days before test; 27 GitHub stars."
],
"docs": [
"pass",
15,
"README over 3,000 chars; install instructions present; `mcpServers` config example present. Check of the package's own README (README shipped in npm package)."
],
"license": [
"pass",
10,
"MIT (OSI-approved)."
]
}
}Install
npx -y @launchdarkly/mcp-server start
Pin the version we tested
npx -y @launchdarkly/mcp-server@0.7.2 start
tested 2026-11-11. Newer versions may differ and were not tested by us.
{
"mcpServers": {
"launchdarkly-mcp": {
"command": "npx",
"args": [
"-y",
"@launchdarkly/mcp-server@0.7.2",
"start"
]
}
}
}Tools listed
create-ai-config, create-ai-config-variation, create-experiment, create-experiment-iteration, create-feature-flag, create-metric, delete-ai-config, delete-ai-config-variation, delete-feature-flag, get-ai-config, get-ai-config-targeting, get-ai-config-variation, get-audit-log-entries, get-code-references, get-environments, get-experiment, get-feature-flag, get-flag-status-across-environments, get-metric, list-ai-configs, list-experiments, list-experiments-any-environment, list-feature-flags, list-metrics, update-ai-config, update-ai-config-targeting, update-ai-config-variation, update-experiment, update-feature-flag, update-metric
Tool history: captured Nov 11, 2026. All tool history.
Quick config
{
"mcpServers": {
"launchdarkly-mcp": {
"command": "npx",
"args": [
"-y",
"@launchdarkly/mcp-server",
"start"
]
}
}
}Add to config Opens the config generator with this server selected.
Setup
- Install. Run
npm install @launchdarkly/mcp-server(Node 20). This is the exact command we ran on 2026-11-11. - Configure. Add the mcpServers entry below to your MCP client config.
- Verify. Restart the client and check that tools/list returns tools (30 in our test).
{
"mcpServers": {
"launchdarkly-mcp": {
"command": "npx",
"args": [
"-y",
"@launchdarkly/mcp-server",
"start"
]
}
}
}Test history
- 2026-11-11: New
FAQ
Is LaunchDarkly MCP deprecated?
No. @launchdarkly/mcp-server was not deprecated or archived when we last tested it on 2026-11-11.
What replaces LaunchDarkly MCP?
Nothing; LaunchDarkly MCP (official) is not deprecated as of 2026-11-11.
Does LaunchDarkly MCP need an API key?
It connects and lists 30 tools without a key, but its tool calls need one.
Embed the badge
Every tested server gets a free embeddable badge. It shows the current score and test date and expires after 90 days. Testing is free; payment never changes a score.