Skip to main content

Playrunner MCP server

Playrunner Cloud speaks the Model Context Protocol, so an AI agent can discover your workflows, start a run, and read the result — without leaving whatever tool you already work in.

Ask an agent to "run the regression suite against staging" and it runs, shards, and reports back.

The endpoint​

https://playrunner.cloud/mcp

One URL, every host. It is a remote server over streamable HTTP, so nothing is installed locally.

Authorisation​

The server uses OAuth 2.1 with PKCE and supports dynamic client registration, which means most hosts need nothing but the URL. Your agent registers itself, sends you to Playrunner to approve, and stores its own token.

You approve a named set of permissions and can revoke them at any time from your Playrunner Cloud account. An agent only ever sees the projects and workflows your account owns.

Add it to a host​

Claude Code​

claude mcp add --transport http playrunner https://playrunner.cloud/mcp

Then run /mcp and authorise when prompted.

Codex​

Playrunner is packaged as a Codex plugin. Install it from the Codex plugin directory, then ask Codex to work with Playrunner — it handles the connection and browser sign-in itself. You do not need an API token or any manual MCP configuration.

The plugin also carries a skill covering workflow authoring, GitHub connection, and the local browser authentication companion used for Authentication Profiles.

note

The plugin listing is still in review. Until it is published, add https://playrunner.cloud/mcp to Codex as a remote MCP server and authorise it in the browser — the same tools are available either way.

VS Code​

Add the server with MCP: Add Server from the command palette, or commit a .vscode/mcp.json so the whole team picks it up:

{
  "servers": {
    "playrunner": {
      "type": "http",
      "url": "https://playrunner.cloud/mcp"
    }
  }
}

Copilot prompts you to authorise in the browser the first time it connects.

Cursor​

Add the server under Settings → MCP, or commit a .cursor/mcp.json to share it with the project:

{
  "mcpServers": {
    "playrunner": {
      "url": "https://playrunner.cloud/mcp"
    }
  }
}

Cursor opens the browser to authorise on first use.

Kody​

  1. Open /account/mcp-servers and choose Add any remote MCP server.
  2. Set the Server name to playrunner and the Server URL to https://playrunner.cloud/mcp.
  3. Kody returns an authorisation link. Open it, sign in to Playrunner Cloud, and approve.
  4. The tools appear as kody.mcp["playrunner"].list_workflows(...).

The name must be lowercase kebab-case — Kody uses it as the accessor in code.

Other hosts​

Anything that accepts a remote MCP server takes the same URL. Where a host asks for a transport, choose streamable HTTP.

What an agent can do​

AreaTools
Discoveryget_account, list_projects, list_workflows, get_workflow, get_authoring_guide
Authoringcreate_project, save_workflow, list_environments, configure_environment
Runningrun_workflow, get_run_status, list_runs
Repositoriesconnect_github, list_github_repositories, list_github_branches, list_integrations
Authenticated testslist_authentication_profiles, save_authentication_profile, authenticate_profile, get_authentication_session, list_companion_devices

run_workflow returns a run id immediately rather than waiting, so an agent starts a suite and polls get_run_status for the outcome.

Limits worth knowing​

Runs started through MCP consume the same workflow-run allowance as any other trigger, and the same per-minute request limits apply. Repeated calls carrying the same idempotency key return the original run instead of starting a second one.

Scoping access​

Give an agent less than your whole account where that matters:

  • Kody can lock a server to specific packages, so ad hoc code cannot call it.
  • Revoke an agent's access from your Playrunner Cloud account whenever you want. Nothing else is affected.
  • Playrunner CLI — the same operations from a terminal or CI
  • Webhooks — trigger a workflow over plain HTTP