Skip to main content
Torchrunner’s surfaces are not the only way in. The workspace exposes an MCP endpoint, so an agent you already work with, Claude Desktop, Claude Code, Cursor, or anything that speaks MCP, can read and write your workspace directly. Your practice’s brain, tasks, notes, and records become tools your agent can call.

​
How it connects

The endpoint lives at: 
https://app.torchrunner.ai/api/mcp
Authentication is a bearer token you issue yourself, inside the app, under Runners → MCP Tokens. Tokens are issued per persona, so an external session acts as a named member of the team rather than as an anonymous key, and they are revocable individually.

​
What an agent can do

The bridge exposes the workspace as tools, discoverable at runtime by any MCP client. The broad strokes:

Brain

Read and write workspace memories, so an external session benefits from and contributes to the same compounding context.

Tasks

Read the board, create and update tasks, delegate work.

Capture

Drop work product from an agent session into the Desk capture inbox as a real, processable document instead of letting it die in chat history.

Records and comms

Query workspace data read-only, and reach connected lanes like email and files where you have linked them.

​
Why this matters

Operators increasingly have their own agent workflows outside any one product. Most software ignores that. Torchrunner treats it as a first-class door: the artifact your Claude session produced at midnight can land on your Desk, filed and triagable, instead of being pasted out of a chat transcript the next morning. One workspace, reachable by every agent you trust, with you controlling the keys.
Tokens carry write access to your workspace. Issue them deliberately, scope them to the persona doing the work, and revoke any token you are not actively using.