Skip to content

Build for NimbleBrain

What you build for NimbleBrain is an MCP server. NimbleBrain connects to it by URL, the agent calls its tools, and, if it ships an interface, people use that interface inside NimbleBrain.

A connector An app
What it is A server alone A server plus its interface
Who uses it The agent, through tools The agent through tools, and people through views
Example Search a ticket system, send an email A task board the agent and the team both edit
You write Python Python, and a web interface

Start with a connector. An app is a connector with an interface added, so nothing is thrown away.

  1. Start from the template

    The Python server template is a working server with tests, lint, CI and packaging set up. Use it, or start from an empty directory with the Server SDK quickstart.

  2. Build the server with the Server SDK

    nimblebrain-server-sdk gives the server what NimbleBrain expects: annotated tools, an error contract, skills, and the ai.nimblebrain extensions. A connector is complete at this step.

  3. Add an interface with Synapse (an app only)

    The interface is a web app built with @nimblebrain/synapse, which gives it a typed way to call the server’s tools. The server serves it as a ui:// resource with FastMCP’s AppConfig; no extra server package is needed. Both work in any MCP Apps host, so Synapse’s documentation is its own site: start with Serve a Synapse UI from a FastMCP server, and see ChatGPT compatibility for that host. What NimbleBrain adds for a view (placements, theming, the bridge) is under App Views.

  4. Publish on mpak

    mpak is the registry for MCP server bundles. Publishing gives the server a versioned, scanned bundle others can install.

  5. Connect it to NimbleBrain

    NimbleBrain connects to a running server by URL. For development, point a workspace at it. For a deployment, an operator adds an entry to the connectors catalog, and the entry’s ai.nimblebrain/host block asks for pages, settings sections, webhooks and notifications.

Everything NimbleBrain expects of a server is a documented protocol. Each protocol page has a “Wire (any SDK)” tab with what to send and read: extensions, host integration and skills. The TypeScript server template is a starting point.