KeelenFor people who work in Claude Code, Cursor, Windsurf, or the Codex CLI
Add one MCP server and tell your agent to set you up.
Updated 12 September 2026
Keelen runs a hosted MCP server at keelen.ai/mcp. Add it to Claude Code, Claude Desktop, Cursor, Windsurf, or the Codex CLI. One line, no key. Signup and email verification are tools of their own. Your agent guides account setup. You confirm your email in the browser. You also complete AI credential connection and GitHub approval there. Then submit a Request from chat. Return to the work and to decisions as they come. Development and merge follow the connected project's settings.
claude mcp add --transport http keelen https://keelen.ai/mcp
Watch it once
An agent adds the server and signs up. It connects a repository, and the first pull request lands. 50 seconds, unstaged.
Claude Code MCP setup demo · August 2026 · 50 seconds
What the 48 tools cover
Your assistant can submit development requests and check roadmap progress. It calls submit_request, get_request_status, list_roadmap and project_status.
Two tools work with no key at all. They are signup and verify_email. The onboarding tools connect GitHub and list repos. They import a project, and they read provisioning and billing. The project tools create projects and submit requests. They answer the planner's questions and reorder the roadmap. They read status, settle escalations, and pause or resume the scheduler. They also list, approve and retire the project's steering rules. One more runs the security scan. That scan is a whole repo security review, and it is not a penetration test.
Same pipeline as the dashboard
A request sent over MCP enters the same intake path as the dashboard. The planner turns it into tasks with acceptance criteria. Development runs build them in single use VMs. A run works with Claude Code, Codex, GLM, Kimi or Grok. Use your own AI subscription or API key, with the methods each provider supports. Another model reviews the diff. A verified test command enables test proof and a clean checkout run. Required repo checks and the project merge policy decide when work can land.
What the server cannot do until you say so
Connecting tokenless only lets your agent sign up and verify an email. It cannot see or change a repo. Connecting GitHub takes an explicit App install that you approve in your browser. Every key from the flow can be revoked in the dashboard. A revoke takes effect at once.
Trigger recurring project reviews from your own scheduler
An external agent or scheduler can call Keelen's MCP tools for a project. The tools are run_security_review, run_legal_exposure_review or run_control_gap_review. You set the cadence outside Keelen, such as daily, weekly or after a release. First complete the scan's required project context. Then handle refused or already running results before you try again. These calls start reviews. They do not silently approve findings. They do not turn every finding into a code change.
How it works
Run the line
In your client. Other clients take the same URL in their own configuration.
Tell your agent to sign you up
It asks for your email and for the six digit code that lands in your inbox.
Say what to build
Approve the provider and GitHub in the browser. Then ask your agent to connect the repo and submit the request.
Ask for progress and decisions
Read what changed and answer questions when the work needs you.
Not for you if
- You want a dashboard. Keelen has one, and this page is about not needing it.
- You do not run an MCP client.
- Your repo has no tests and you will not add any.
- You want the agent in your terminal to do the work itself. It delegates to the loop, and the loop works in its own VMs.
Paste the line. Say set me up.
Signing up, connecting GitHub, and creating a project cost nothing. Running the loop is a 14 day trial with no card, then flat tiers from $29 a month plus your own coding model usage. Keelen pays for chat.