{
  "markdown": "# launchflag-mcp\n\nMCP server for [LaunchFlag](https://launchflag.dev). It lets a coding agent create a feature flag, ship the\nrisky path dark behind it, and flip it later without a deploy.\n\nSeven tools, one token, no state.\n\nYou probably do not need this package. `launchflag` is the SDK and the CLI, and the hosted MCP server needs\nnothing installed. `launchflag-mcp` is only for running that same server locally instead.\n\n## Start here\n\n```bash\nnpx launchflag init --key lf_dev_… --token lfk_… --url https://launchflag.dev\n```\n\nFrom your project root. It installs `launchflag`, writes the key, downloads the agent skill and points your\nclient at the hosted MCP server.\n\n## Hosted, or local\n\n**Hosted (the default).** Nothing to install. Point your client at `https://launchflag.dev/mcp` with your API\ntoken (`lfk_…`) in the `Authorization` header.\n\n**Local stdio (the alternative).** Runs on your machine and talks to the same API over HTTPS. Useful when your\nclient does not speak Streamable HTTP, or when you self-host LaunchFlag.\n\n```bash\nnpx -y launchflag-mcp@latest\n```\n\n## Env vars\n\n| Variable | Required | Default | What it is |\n| --- | --- | --- | --- |\n| `LAUNCHFLAG_TOKEN` | yes | none | Your API token (`lfk_…`), from the dashboard Install page. The server exits if it is missing. |\n| `LAUNCHFLAG_URL` | no | `https://launchflag.dev` | Base URL of the LaunchFlag instance. Set it only if you self-host. |\n\nThe token is a secret. Keep the file that holds it out of git.\n\n## The seven tools\n\n| Tool | What it does |\n| --- | --- |\n| `list_flags` | Every flag with its dev, staging and prod state, fail mode and variants. |\n| `create_flag` | Creates a flag OFF in all three environments and returns the snippet to wrap the new path in. |\n| `set_flag` | Turns a flag on or off, or changes who it targets, in one environment. |\n| `promote_flag` | Copies one environment's config forward: dev to staging, or staging to prod. |\n| `delete_flag` | Deletes a flag, once the code paths behind it are gone. |\n| `emergency_off` | Kill switch: disables every non-essential flag in the project at once, or lifts it. |\n| `list_projects` | Your projects and their env keys. |\n\n## Claude Code\n\nHosted:\n\n```bash\nclaude mcp add --transport http launchflag https://launchflag.dev/mcp --header \"Authorization: Bearer lfk_…\"\n```\n\nLocal, in `.mcp.json`:\n\n```json\n{\n  \"mcpServers\": {\n    \"launchflag\": {\n      \"command\": \"npx\",\n      \"args\": [\"-y\", \"launchflag-mcp@latest\"],\n      \"env\": { \"LAUNCHFLAG_TOKEN\": \"lfk_…\" }\n    }\n  }\n}\n```\n\n## Cursor\n\n`.cursor/mcp.json`, hosted:\n\n```json\n{\n  \"mcpServers\": {\n    \"launchflag\": {\n      \"type\": \"http\",\n      \"url\": \"https://launchflag.dev/mcp\",\n      \"headers\": { \"Authorization\": \"Bearer lfk_…\" }\n    }\n  }\n}\n```\n\nSwap in the `command`/`args`/`env` block above to run it locally instead.\n\n## Codex\n\n`codex mcp add` has no header flag: it reads a bearer token from an environment variable, which it\nre-reads every time Codex starts, so put the export in your shell profile.\n\n```bash\nexport LAUNCHFLAG_TOKEN=lfk_…\ncodex mcp add launchflag --url https://launchflag.dev/mcp --bearer-token-env-var LAUNCHFLAG_TOKEN\n```\n\nLocal:\n\n```bash\ncodex mcp add launchflag --env LAUNCHFLAG_TOKEN=lfk_… -- npx -y launchflag-mcp@latest\n```\n\n## Docs\n\nFull agent guide, including the skill file and the flag conventions agents should follow:\n[launchflag.dev/docs/agents](https://launchflag.dev/docs/agents).\n\nQuestions or a bug: [contact@launchflag.dev](mailto:contact@launchflag.dev).\n\nMIT licensed.",
  "bytes": 3512,
  "sha": "52a5b3653df88f9ab7badb52cd2d646b13a6f9b61e575c7a14f6f38015060f86",
  "repo_slug": "adarshmishra07/flagship",
  "fonte": "npm",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_dev_launchflag_launchflag_a7df1c87/readme"
}