Octavus MCP Server
The Octavus MCP server lets you connect your Octavus account to an AI coding tool - Cursor, Claude Code, or any MCP-compliant client - and work with your Octavus resources without leaving your editor.
Once connected, your tool can:
- Inspect and debug platform agents - list your projects and agents, read an agent's full definition, and read a session's execution trace to see exactly what happened during a run.
- Deploy changes to platform agents - edit an agent's files locally and push them back, the same way the
octavusCLI syncs. - Read and drive your Octavus Agents - list your agents and their chat threads, read a thread, and send a new message ("Ask Anna BDR to draft the follow-up email"), then read the response.
How it works
The MCP server is a remote HTTP server - there is no package to install and no local process to run. You point your client at one URL and sign in with your Octavus account through OAuth:
- Your client discovers how to authenticate from the server URL.
- You sign in to Octavus (Google or email/password) and approve the connection.
- You pick an access level - read-only or read and write.
- Your client is connected and stays connected; you never copy or paste a token.
Acting as you
Every action the MCP server takes runs as you, with exactly the permissions you have in the Octavus dashboard - the same projects, agents, and Octavus Agents you can see and manage. It can never do more than you can, and a read-only connection can never make changes. The connection is scoped to a single organization.
You can review and disconnect connected apps at any time from Account -> Connected apps.
Environments
| Environment | MCP URL |
|---|---|
| Production | https://octavus.ai/mcp |
| Staging | https://octavus.dev/mcp |
Continue to Connecting to add the server to your client.