B Buffy Agent
Buffy Agent Blog · Engineering

MCP Tools With the Buffy API: Use `/mcp` for Real Automations

How to use Buffy’s MCP endpoint to list and invoke tools tied to your user, with a practical “first automation” workflow.

MCP Tools With the Buffy API: Use /mcp for Real Automations

Once you’ve wired the Buffy behavior core into your app, the next question is usually:

How do you connect that behavior to the tools your automation agents already use?

Buffy’s API exposes an MCP endpoint you can use to list and invoke tools backed by your user identity.

This post gives you a “first week” workflow for using /mcp without guesswork.

What you’ll learn

  • What /mcp is for in the Buffy integration story
  • A simple automation flow: list tools → pick one → invoke it → verify impact
  • Where MCP fits vs raw REST calls (habits/events) in your product design

Start from the reference overview if you want the exact shapes:

What /mcp gives you

Think of REST vs MCP like this:

  • REST calls are your “direct API” surface (messages, activities, events).
  • MCP tools are your “agent tool” surface: you can let an automation agent call specific actions as tools.

In practice, /mcp lets you:

  • discover tools available for your user
  • invoke them with the right Authorization identity
  • keep tool actions grounded in the same Buffy behavior core (so state stays coherent)

First automation: list tools, invoke one, confirm state

Step 1: authenticate

Buffy uses Bearer auth for MCP as well:

Authorization: Bearer <YOUR_API_KEY>

Step 2: list available tools

In your integration, call /mcp and ask for the tool inventory exposed by your account. (OpenAPI provides the exact request body shape; treat that as canonical.)

Step 3: pick one tool and invoke it

Choose a tool that produces an obvious state change in Buffy, such as:

  • logging a habit completion
  • requesting today’s scheduled activities
  • triggering a routine or follow-up flow

Then invoke it from your automation agent.

Step 4: verify the behavioral impact

After the tool runs, confirm what changed by querying:

  • today’s activities
  • the recent event history

This keeps your automation debuggable and prevents “tool did something, but nothing changed” confusion.

If you want a developer-friendly baseline for those calls:

Where MCP fits in the overall integration model

MCP tools are an adapter layer. The product story stays the same:

  • messages and tool invocations update the behavior core
  • the core updates activities, reminders, and history
  • your app/agent decides what the user sees next

If you’re building multi-agent systems, this keeps your behavior consistent across surfaces.

Background:

Next step

If you want to move from “calling tools” to “full loop UX”, start with:

Further reading