LaunchDarkly MCP Server

Official

Local MCP server for LaunchDarkly Federal and EU environments

How to install ↓ Compare GitHub source repository ↗

Category: Dev Tools #280 of 438
Stars: ★ 27 Popular
Transport: stdio (local process) · SSE
Runtime: Node.js · Docker · Prebuilt binary
Credentials: API key / credential required
License: MIT
Last commit: 20 days ago
Tools: 16

48FMRS · D

The LaunchDarkly local MCP server is a self-managed solution for Federal and EU environments with limited features compared to the hosted version. It requires an API key and allows AI clients to interact with LaunchDarkly data via stdio or SSE. Suitable for users who cannot use the hosted MCP, but be aware of security risks and feature limitations.

Strongest · Setup experience 11/20 Weakest · Reliability 8/20

Reliability

8/20

Security and permissions

8/20

Maintenance

10/20

Documentation

11/20

Setup experience

11/20

Why each score

Reliability 8/20: All evidence is limited to the README and repository metadata; there is no CI workflow, no committed test suite, and no cached server manifest, so a clean MCP init + tool-list handshake cannot be verified. The README states the library is programmatically generated and that manual internal changes will be overwritten, yet it also documents local-clone builds (npm install / npm run build) and multi-platform Bun binaries, leaving the build/release consistency questionable. A large tool surface is declared (AI Configs, FeatureFlags, AuditLog, Environments, CodeReferences) but with no parameter details and no error-handling documentation (several README sections are empty placeholders), so declared tools cannot be matched against real behavior. Static calibration caps this dimension at 12 without execution evidence; additionally, the Docker install snippet contains a malformed mcpServers JSON (missing comma after the env object before url), a visible contradiction that would break that example, so the score is set to 8.
Security and permissions 8/20: No red line is triggered: tokens in the installation examples are obvious placeholders (api-xxxxxxxx-...), not real secrets, and there is no sign of credential theft, covert exfiltration, or malicious code. However, several concrete deductions apply. First, every install path passes a long-lived LaunchDarkly API key in plaintext via --api-key on the command line, which leaks into process listings, shell history, and client config files, and no least-privilege key guidance (e.g. read-only tokens) is given. Second, the Docker example is internally inconsistent (--api-key "$LD_ACCESS_TOKEN" alongside an env mapping of LD_ACCESS_TOKEN to MCP_LD_TOKEN), which is misleading about how credentials flow. Third, the tool set includes destructive and write operations (delete, patch, updateTargeting) with no stated confirmation step, scope limits, or data-flow disclosure. Fourth, storage and logging behavior of the self-managed server's API key is undocumented. Least-privilege and dangerous-operation confirmation are both incomplete, so 8 rather than lower (no red line) or higher (security boundaries not closed).
Maintenance 10/20: Positives: an explicit MIT license, a non-archived repository, topics including managed-by-terraform and launchdarkly-integration indicating organizational ownership and infrastructure management, a well-known commercial vendor, and links to official hosted-MCP and authorization documentation. Deductions: in a static snapshot there is no visibility into commit history, release cadence, dependency updates, or issue response time; only 23 stars and 15 open issues are available, and popularity signals must not add points. The README says the library is programmatically generated and that manual changes to internal files will be overwritten, which effectively constrains the contribution path, and no versioning or governance policy is documented beyond a {tag} placeholder in a releases URL. No security vulnerability reporting channel (e.g. SECURITY.md) is visible and the Authentication section is empty, so the security-response path is unclear. Active but with governance and response gaps, hence 10.
Documentation 11/20: Documentation is reasonably layered: it covers seven installation routes (Cursor, Claude Desktop, Qodo Gen, GitHub Copilot CLI, standalone binary, local clone, environment variables, and AWS Marketplace Docker), and supplies a server-url table for Commercial, Federal, and EU environments. It also candidly discloses a real limitation: the hosted MCP server is more feature-complete and updated more frequently than this self-managed one, which is a genuine and useful caveat. Deductions are substantial: the Authentication section is empty, with no guidance on key creation, scopes, or rotation; the tool list gives only method names and one-line summaries with no parameters, responses, pagination, or rate limits; sections for usage, errors, retries, debugging, and http-client are empty placeholders; the docs/sdks/* links are relative and unverifiable in this snapshot; the Docker configuration example has a JSON syntax error and a self-contradictory env/--api-key mapping that would directly break setup, and there is no troubleshooting section at all. Usable but with hidden assumptions and troubleshooting gaps, so 11.
Setup experience 11/20: The entry path is relatively friendly: a single npx -y --package @launchdarkly/mcp-server command, plus copy-paste mcp.json / claude_desktop_config.json snippets for mainstream clients such as Cursor, Claude Desktop, and Copilot CLI. Alternatives are offered too (dependency-free standalone binary, local clone build, and Docker), and switching environments is just an added --server-url flag. Static calibration caps this at 15 absent CI or test evidence, and real friction remains: the four install snippets are near-duplicates and verbose; the API key must be replaced by hand; the local-clone path requires manual npm install and npm run build; the README itself admits the env-var approach fails in clients like Cursor that do not support environment variables within MCP config, requiring a local-script workaround; the Docker example is missing a JSON comma and has a contradictory env mapping, so copying it verbatim fails; and the local-clone example uses an inconsistent lowercase server name (launchdarkly). Clear but manual and with defective examples, hence 11.

Static review · not runListed 2026-09-10

Read the FMRS scoring method →

Fit and risk

What it can accessUses the networkChanges third-party account data

Best for

Teams using LaunchDarkly Federal or EU environments
Developers who want to manage feature flags via AI clients

Not for

Users who need the full feature set of the hosted MCP
Users unwilling to provide a LaunchDarkly API key

Required permissions

Requires a LaunchDarkly API key for authentication
Accesses feature flags, AI Configs, environments, audit logs, and code references data in your LaunchDarkly account

Risks and side effects

API key exposure could lead to unauthorized access
Modifications to feature flags may affect production environments
The server is self-managed and may lack features or timely updates

Setup

Before you start

Runtime：Node.js · Docker · Prebuilt binary

LD_ACCESS_TOKEN optionalsecret LaunchDarkly API access token, generated on LaunchDarkly's Authorization page.

MCP_LD_TOKEN optionalsecret An API token supplied via environment variable, referenced through the $LD_ACCESS_TOKEN placeholder to inject real credentials.

For Cursor, create a .cursor/mcp.json file in your project root with the content {"mcpServers":{"LaunchDarkly":{"command":"npx","args":["-y","--package","@launchdarkly/mcp-server","--","mcp","start","--api-key","api-xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"]}}}, replacing the api-key with your LaunchDarkly API key. For Claude Desktop, add the same content to claude_desktop_config.json. For Qodo Gen, click 'Connect more tools', then '+ Add new MCP', add the same configuration and save. For GitHub Copilot CLI, run /mcp add or edit ~/.copilot/mcp-config.json. Alternatively, install via standalone binary, local clone, or Docker.

.cursor/mcp.json

{"mcpServers":{"LaunchDarkly":{"command":"npx","args":["-y","--package","@launchdarkly/mcp-server","--","mcp","start","--api-key","api-xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"]}}}

Shown for Cursor. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json

{
  "servers": {
    "LaunchDarkly": {
      "command": "npx",
      "args": [
        "-y",
        "--package",
        "@launchdarkly/mcp-server",
        "--",
        "mcp",
        "start",
        "--api-key",
        "api-xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal

claude mcp add LaunchDarkly -- npx -y --package @launchdarkly/mcp-server -- mcp start --api-key api-xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After installation, check the client's tool list for tools like getTargeting, list, and getStatus; then ask "list my feature flags" — a returned list of LaunchDarkly flags proves the API key and connection work.

Troubleshooting

Ensure the API key is valid and has permissions to access the required resources
Verify that the --server-url argument matches your environment (commercial, Federal, or EU)
Check network connectivity and that the AI client's MCP configuration format is correct
For clients using environment variables, confirm that variable names and values are passed correctly

Things to try

Once connected, you can ask your AI assistant things like:

List all feature flags in this project
Show the status of the enable-new-checkout flag across environments
Create a new feature flag called trial-banner
List recent audit log entries

Tools 16

getTargeting read-only

Show an AI Config's targeting

updateTargeting writes

Update AI Config targeting

list read-only

List AI Configs

create writes

Create new AI Config

delete destructive

Delete AI Config

get read-only

Get AI Config

update writes

Update AI Config

createVariation writes

Create AI Config variation

Show 8 more tools

deleteVariation destructive

Delete AI Config variation

getVariation read-only

Get AI Config variation

updateVariation writes

Update AI Config variation

listEntries read-only

List audit log entries

listRepositories read-only

List code references repositories

listByProject read-only

List environments by project

getStatus read-only

Get flag status across environments

patch writes

Update feature flag

Use cases

Manage LaunchDarkly feature flags and AI Configs

View flag status across environments

List audit log entries

List code reference repositories

List environment information

Supported clients

Cursor

Claude Desktop

Qodo Gen

GitHub Copilot CLI

Listed from the project's documentation, not tested by this site.

Overview

This is the local Model Context Protocol (MCP) server for LaunchDarkly, specifically designed for LaunchDarkly's Federal and European Union (EU) environments. These environments do not yet support the hosted MCP, so LaunchDarkly provides this self-managed server as an alternative. LaunchDarkly strongly recommends using the hosted server where available, as it is more feature complete and receives more frequent updates. The server can be installed via npx, standalone binary, local clone, or Docker, and allows AI clients to connect via stdio or through a URL using SSE. It requires a LaunchDarkly API key and supports specifying the commercial, Federal, or EU environment via the --server-url argument.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →

NPM Sentinel MCP Server 79 · B

AI-powered NPM package analysis MCP server

★ 18 · Tools 19 Compare with this →

MartinLoop MCP Server 79 · B

An independent control layer that gives AI coding agents budgets, brakes, and receipts

★ 413 Compare with this →

GitHub MCP Server 76 · B

GitHub's official MCP server for managing repos, issues, PRs, and workflows via natural language

★ 33.7k · Tools 15 Compare with this →

Compare with other MCP servers

Source revision c3d39cfb0858 Data synced 2026-10-19 Read the FMRS scoring method