What CodeLynq does with MCP
CodeLynq exposes an MCP endpoint, so an AI assistant you already use can drive it directly. Fifteen tools cover the pipeline: listing projects, repositories and branches, reading the architecture map, creating a task, generating and approving a technical spec, starting a run and checking its status, generating documentation, listing releases, triggering a deployment, and opening a troubleshooting session against a real environment.
In practice that means you can stay in the assistant you are already talking to. Describe the work there, let it file the task and start the run, and review the pull request in your normal queue.
Setting it up
- Create an API token in CodeLynq, or connect through the OAuth flow if your client supports it.
- Point your MCP client at the CodeLynq endpoint and authenticate.
- Ask the assistant to list your projects to confirm the connection.
- From there, create a task and take it through spec approval to a run.
Worth knowing
Access is scoped, not blanket. A read-only token can list and inspect but cannot start anything, and each tool declares whether it writes. The OAuth server implements the current spec properly — PKCE is mandatory, plain challenges are rejected, redirect URIs match exactly rather than by prefix, and authorisation codes are single-use. Every action still lands in the audit log attributed to you, so "an assistant did it" is never the end of the trail.
Other integrations
CodeLynq also works with coding agents, GitHub, GitLab, Bitbucket, GitHub Actions, Fireflies, Plaud, Slack and webhooks.