CircleCI

Diagnose failed builds, read job logs, inspect test results and rerun or roll back deployments from your AI assistant.

Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.

Paste it into any MCP client. Setup docs

What the CircleCI connector does

The CircleCI MCP server connects your CI/CD pipelines to Claude, ChatGPT and any MCP-compatible agent. It exposes 24 tools for finding out why a build failed, reading job logs and test results, listing build artifacts, and rerunning, cancelling or rolling back work that has already shipped. CircleCI hosts it, so nothing installs locally.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on .

Anthropic states this reflects the level of review a connector received, not a security audit.

CircleCI tools (24)

  • cancel_workflow
  • download_usage_data
  • get_deploy_component
  • get_deploy_environment
  • get_job
  • get_job_logs
  • get_job_resource_usage
  • get_me
  • get_orb
  • get_orb_source
  • get_run
  • get_workflow
  • list_deploy_component_versions
  • list_deploy_components
  • list_deploy_environments
  • list_deployments
  • list_job_artifacts
  • list_job_tests
  • list_run_workflows
  • list_runs
  • list_workflow_jobs
  • rerun_workflow
  • rollback_deploy_component
  • validate_config

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

  • The hosted server is in preview. CircleCI states plainly that you can expect changes to the available tools during this phase, and points users at its community forum for feedback. Do not build anything load-bearing on a specific tool name.
  • Rollbacks require prior setup. Deploy markers must be configured, the code must live in a GitHub repository, and the CircleCI GitHub App must be installed in the organisation. Without those, rollback_deploy_component has nothing to act on.
  • download_usage_data caps at a 31-day range per call, per CircleCI's tool reference. Longer reporting periods need multiple calls.
  • API rate limits apply, and they are not generous for polling. CircleCI's API v2 index states a limit of 5,000 requests per hour per token, and its developer guide warns that the API protects against sudden bursts and sustained high volumes such as frequent polling. Throttled requests return HTTP 429, usually with a Retry-After header. The MCP endpoint itself returned ratelimit-limit: 50 per second in our probe on 2026-09-08.
  • No config authoring, orb publishing, context or environment-variable management. CircleCI assigns all of that to the CLI MCP, not the hosted server. If you need it, connect the CLI MCP as well.
  • Read-only tools cannot fix anything. The server surfaces logs, tests and artifacts; applying a fix still means editing your repository. The only writes available are rerun, cancel and rollback.
  • We could not read the server's tool annotations. The endpoint requires authentication, so the read/write classification on this page rests on CircleCI's documentation rather than on MCP safety annotations from the server.
  • Tool counts disagree between sources. Anthropic's directory lists 24 tools where CircleCI's documentation table lists 13, with several names differing. See the comparison above.

Frequently asked questions

Does the CircleCI MCP server need a paid plan?

No. CircleCI's connection guide lists only a CircleCI account with access to the projects you want to inspect, plus an MCP host that supports remote servers over streamable HTTP. Rollback tooling is the exception: CircleCI's rollback setup guide requires a GitHub repository, the CircleCI GitHub App, and configured deploy markers before any rollback can run.

Is the CircleCI MCP server OAuth or an API token?

Both, and OAuth2 is the recommended path. CircleCI's docs say most MCP hosts detect the OAuth2 requirement and prompt you to sign in on first connect. A personal API token sent as a bearer header is the documented fallback, intended only for hosts that cannot complete an OAuth2 flow, or for headless use.

Can the CircleCI MCP server cancel or roll back a real deployment?

Yes. Four of the 24 tools change CI/CD state: cancel_workflow, rerun_workflow, rollback_deploy_component and validate_config, of which the first three act on live infrastructure. CircleCI's own best-practice guidance says to review destructive actions before confirming them, noting this matters most on a shared or production-facing project.

Why does my assistant not see the newest CircleCI tools?

Restart your MCP host or reconnect the server. CircleCI's troubleshooting section states that after it releases new tools, an MCP host must be restarted or the server reconnected before the updated list loads. The hosted server is in preview, so CircleCI warns that the available tools will change during this phase.

What is the difference between the hosted CircleCI MCP server and the CircleCI CLI MCP?

The hosted server runs at mcp.circleci.com and exposes a curated set of tools for observing and acting on CI runs. The CLI MCP runs locally and exposes the entire CircleCI CLI, including config authoring, orb tooling, policy and runner management. CircleCI recommends the hosted server for day-to-day diagnostic work.

Is my CircleCI API token shared with Claude?

Only if you choose the token fallback, and CircleCI advises against storing it in plain config. Its token guide notes that manually created personal API tokens grant full read and write permissions, while tokens issued through the OAuth flow are scoped to the Read, Write or Admin level the user picks at authorization time.

Sources

Use it in an agent

Put CircleCI to work.

Connect CircleCI once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.

Call (650) 285-1019Our AI receptionist answers.