KeelenAnswer — MCP
Can I run a whole project from Claude or another AI chat?
Updated 20 August 2026
Yes, through the Model Context Protocol. MCP lets an AI client call external tools, so your agent can do more than read context: it can act. Most MCP servers today supply context (files, databases, browsers, issue trackers). A smaller number do work. Keelen's hosted server is the second kind: register it, say "set me up", and your agent creates the account, connects your GitHub repository, and submits build requests. From then on you steer an autonomous dev loop from the chat window, and the work comes back as tested, merged pull requests.
Register it in Claude Code with a single command: claude mcp add --transport http keelen https://keelen.ai/mcp
What to do, in order
Pick a client that supports MCP
Claude Code, Claude Desktop, Codex, Cursor, and Windsurf all do. The protocol is the same across them, so the server works wherever you already are.
Register the server
For Claude Code that is one command: claude mcp add --transport http keelen https://keelen.ai/mcp. Other clients take the same URL in their own configuration.
Let the agent do the setup
Hand it the setup instructions and say "set me up". It asks for your email, asks for the code that lands in your inbox, then handles engine connection, GitHub connection, and the first project.
Steer in plain language
Submit requests, reorder the roadmap, check status, and read the pull requests that come back, all from the chat window. The web dashboard is optional.
Keep the gates on
Driving from chat changes the interface, not the safety model. Tests, review, and CI still stand between a request and your main branch.
Context servers and working servers
Most of the MCP ecosystem gives an agent things to read: filesystems, databases, browsers, design files, issue trackers. That is useful and it is where the popular lists concentrate. A server that accepts a goal and returns shipped work is a different shape, because the work outlives the conversation: the loop keeps running after you close the window.
What happens after you close the chat
The request you submitted becomes roadmap items and dev-ready tasks. The loop implements them in isolated VMs on your own credentials, one pull request at a time, and keeps going on its own schedule. You can check in from any client later, or from the dashboard, and pause it whenever you like.
The safety model does not change
Five gates stand between the loop and your main branch. New tests are applied without the implementation first and must fail, so tests born green are rejected. An independent reviewer that shares no context with the run that wrote the code reviews the diff. Your own test suite runs on a clean checkout. Nothing merges over a red required CI check. And a review window stands before gated auto-merge, which you can replace with plan-review sign-off, manual merge, or branch-only mode.
Where the code runs
Every iteration runs in its own single-use, non-root VM that is destroyed when the run ends. Network egress is deny-by-default through a proxy. The GitHub token an iteration holds is scoped to one repository for about an hour and cannot modify your CI workflows.
When Keelen is not the answer
- Your client does not support MCP. The dashboard does everything the server does.
- You want the agent to edit files in your local checkout directly. That is what a CLI coding agent is for; Keelen works on the repository in its own isolated VMs.
- You are not comfortable connecting a GitHub repository to a hosted service.
Hand the work to a loop
Connect a repository, write what you want in plain language, and review the tested pull requests that come back.
FAQ
What is MCP?
The Model Context Protocol: an open standard that lets AI clients call external tools and data sources through one interface. It is what allows an assistant to act on a system rather than only describe it.
Do I need the web dashboard at all?
No. Signup, GitHub connection, engine connection, requests, and status all work over MCP. The dashboard is there when you want the visual view of the roadmap and metrics.
Which MCP clients work with it?
Any MCP-capable client over HTTP. Per-client setup guides exist for Claude Code, Claude Desktop, Codex, Cursor, Windsurf, and generic clients.