Skip to main content

Command reference

Every example on this page is real output captured from mcp running against a live workspace. Identifiers and key previews are masked (xxxx), but the shape, columns, and values are exactly what the CLI prints.
Global flags work on every command: --json (raw JSON), --format <table|tsv|jsonl|json>, --field <names>, --filter <key=substring>, --profile <name>, and --base-url <url>. Run mcp --help for the full list.

Auth and identity

mcp whoami resolves your identity, the active API key, the base URL, and every organization you can act on:

Projects

Servers

mcp servers list — one row per server with its project, status, and live version:
mcp servers get <serverId> — the deployed URL is the live MCP endpoint:
mcp servers logs <serverId> — recent deployment events (newest first):

Tools

mcp tools list --server <serverId> — every generated tool with its enrichment status:
mcp tools show <name> --server <serverId> surfaces the applied metadata plus any pending AI suggestion:

Deployments

mcp deployments runtime-auth <deploymentId> — the access posture and edge policy for a deployment, including rate/request limits:

Invoke a tool

mcp invoke exchanges runtime-auth if needed, sends a JSON-RPC tools/call, and prints the response — a fast way to smoke-test a deployed server:
--args accepts inline JSON, @path/to/args.json, or @- for stdin. Add --json for the full JSON-RPC envelope, or --raw to skip the content-block pretty printer.

Skills

Marketplace

API keys

Config and profiles

mcp config show prints the active profile (the API key is always redacted):

Shell completion

Diagnostics

Run mcp doctor whenever something looks off — it checks Node, your config, the base URL, API-key auth, the claude CLI, and whether a newer release is available. See Diagnostics & the interactive UI for the full walkthrough.