Skip to main content
The Runtime MCP server lets an AI client create, run and govern your Runtime agents from chat. It calls the same API as the dashboard and the Agent CLI, with the same scopes, roles and rate limits.
Streamable HTTP. Sign in with OAuth (your Runtime login, including SSO) or send a Runtime API key as a bearer token. See Authentication.

Quick start

Pick your client. The first time you use it, it opens a browser to sign in to Runtime.
Then run /mcp in Claude Code, pick runtime and sign in.
Check it works: ask your client to call runtm_whoami. It returns your user, organization, role and scopes.

Use an API key instead

For clients without OAuth, MCP gateways and CI, send a Runtime API key as a bearer token. The key’s scopes decide what the client can do.
Never paste an API key into a chat. Keep it in the client’s config or an environment variable.

What it can do

31 tools load by default. Add or drop groups with ?toolsets=. See Tools.

Safe by design

  • Your permissions, no more. Every call runs through the REST API with your scopes, org role and rate limits. Tools you can’t use are hidden.
  • No credentials in chat. A step that needs a key, token or password returns a dashboard link for you to finish. A request field named like a credential is refused.
  • Sandbox output is data. Session events, command output and files are labelled untrusted, and the model is told not to follow instructions in them.
  • A 403 is final. The model is told to name the missing permission, not to look for another way in.

Next

Authentication

OAuth, workspaces, scopes and API keys.

Tools

Every tool, its inputs and toolsets.

Okta

Sign in with Okta, and connect Okta-protected MCP servers to agents.

Troubleshooting

Fix sign-in, 401, 403 and missing tools.