How to integrate Braintrust MCP with OpenAI Agents SDK

This guide walks you through connecting Braintrust to the OpenAI Agents SDK using the Composio tool router. By the end, you'll have a working Braintrust agent that can compare failed evaluations across recent experiments, create dataset from production failure logs, summarize latency spikes in braintrust traces through natural language commands. This guide will help you understand how to give your OpenAI Agents SDK agent real control over a Braintrust account through Composio's Braintrust MCP server. Before we dive in, let's take a quick look at the key ideas and tools involved.

Braintrust logoBraintrust
Api Key

Braintrust is an AI engineering platform for prompts, datasets, experiments, evaluations, observability, and production AI apps. Use it to ship better AI systems with faster debugging, clearer evals, and tighter feedback loops.

17 Tools

Introduction

This guide walks you through connecting Braintrust to the OpenAI Agents SDK using the Composio tool router. By the end, you'll have a working Braintrust agent that can compare failed evaluations across recent experiments, create dataset from production failure logs, summarize latency spikes in braintrust traces through natural language commands.

This guide will help you understand how to give your OpenAI Agents SDK agent real control over a Braintrust account through Composio's Braintrust MCP server.

Before we dive in, let's take a quick look at the key ideas and tools involved.

Also integrate Braintrust with

TL;DR

Here's what you'll learn:
  • Get and set up your OpenAI and Composio API keys
  • Install the necessary dependencies
  • Initialize Composio and create a Tool Router session for Braintrust
  • Configure an AI agent that can use Braintrust as a tool
  • Run a live chat session where you can ask the agent to perform Braintrust operations

What is OpenAI Agents SDK?

The OpenAI Agents SDK is a lightweight framework for building AI agents that can use tools and maintain conversation state. It provides a simple interface for creating agents with hosted MCP tool support.

Key features include:

  • Hosted MCP Tools: Connect to external services through hosted MCP endpoints
  • SQLite Sessions: Persist conversation history across interactions
  • Simple API: Clean interface with Agent, Runner, and tool configuration
  • Streaming Support: Real-time response streaming for interactive applications

What is the Braintrust MCP server, and what's possible with it?

The Braintrust MCP server is an implementation of the Model Context Protocol that connects your AI agent and assistants like Claude, Cursor, etc directly to your Braintrust account. It provides structured and secure access to your projects, prompts, datasets, experiments, evaluations, and observability data, so your agent can create resources, manage evaluation data, inspect events, query results, and update configurations on your behalf.

  • Project and resource management: Have your agent create, list, update, or delete Braintrust projects and manage the experiments, datasets, and prompts within them.
  • Dataset creation and maintenance: Let the agent create datasets, add batches of examples, revise dataset details, and summarize their contents and metrics.
  • Experiment tracking and evaluation: Direct your agent to create experiments, add result events, update experiment details, and retrieve aggregate evaluation summaries.
  • Prompt configuration: Instruct your agent to create versioned prompts and update their metadata or configuration without running them.
  • Observability and data analysis: Have your agent fetch recent project, experiment, or dataset events and run read-only queries over Braintrust data to investigate results.

What is the Composio tool router, and how does it fit here?

What is Composio SDK?

Composio's Composio SDK helps agents find the right tools for a task at runtime. You can plug in multiple toolkits (like Gmail, HubSpot, and GitHub), and the agent will identify the relevant app and action to complete multi-step workflows. This can reduce token usage and improve the reliability of tool calls. Read more here: Getting started with Composio SDK

The tool router generates a secure MCP URL that your agents can access to perform actions.

How the Composio SDK works

The Composio SDK follows a three-phase workflow:

  1. Discovery: Searches for tools matching your task and returns relevant toolkits with their details.
  2. Authentication: Checks for active connections. If missing, creates an auth config and returns a connection URL via Auth Link.
  3. Execution: Executes the action using the authenticated connection.

Step-by-step Guide

Step by step09 STEPS
1

Prerequisites

Before starting, make sure you have:
  • Composio API Key and OpenAI API Key
  • Primary know-how of OpenAI Agents SDK
  • A live Braintrust project
  • Some knowledge of Python or Typescript
2

Getting API Keys for OpenAI and Composio

OpenAI API Key
  • Go to the OpenAI dashboard and create an API key. You'll need credits to use the models, or you can connect to another model provider.
  • Keep the API key safe.
Composio API Key
3

Install dependencies

npm install @composio/openai-agents @openai/agents dotenv

Install the Composio SDK and the OpenAI Agents SDK.

4

Set up environment variables

bash
OPENAI_API_KEY=sk-...your-api-key
COMPOSIO_API_KEY=your-api-key
USER_ID=composio_user@gmail.com

Create a .env file and add your OpenAI and Composio API keys.

5

Import dependencies

import 'dotenv/config';
import { Composio } from '@composio/core';
import { OpenAIAgentsProvider } from '@composio/openai-agents';
import { Agent, hostedMcpTool, run, OpenAIConversationsSession } from '@openai/agents';
import * as readline from 'readline';
What's happening:
  • You're importing all necessary libraries.
  • The Composio and OpenAIAgentsProvider classes are imported to connect your OpenAI agent to Composio tools like Braintrust.
6

Set up the Composio instance

dotenv.config();

const composioApiKey = process.env.COMPOSIO_API_KEY;
const userId = process.env.USER_ID;

if (!composioApiKey) {
  throw new Error('COMPOSIO_API_KEY is not set. Create a .env file with COMPOSIO_API_KEY=your_key');
}
if (!userId) {
  throw new Error('USER_ID is not set');
}

// Initialize Composio
const composio = new Composio({
  apiKey: composioApiKey,
  provider: new OpenAIAgentsProvider(),
});
What's happening:
  • dotenv.config() loads your .env file so COMPOSIO_API_KEY and USER_ID are available as environment variables.
  • Creating a Composio instance using the API Key and OpenAIAgentsProvider class.
7

Create a Tool Router session

// Create Tool Router session for Braintrust
const session = await composio.create(userId as string, {
  toolkits: ['braintrust'],
});
const mcpUrl = session.mcp.url;

What is happening:

  • You give the Tool Router the user id and the toolkits you want available. Here, it is only braintrust.
  • The router checks the user's Braintrust connection and prepares the MCP endpoint.
  • The returned session.mcp.url is the MCP URL that your agent will use to access Braintrust.
  • This approach keeps things lightweight and lets the agent request Braintrust tools only when needed during the conversation.
8

Configure the agent

// Configure agent with MCP tool
const agent = new Agent({
  name: 'Assistant',
  model: 'gpt-5',
  instructions:
    'You are a helpful assistant that can access Braintrust. Help users perform Braintrust operations through natural language.',
  tools: [
    hostedMcpTool({
      serverLabel: 'tool_router',
      serverUrl: mcpUrl,
      headers: { 'x-api-key': composioApiKey },
      requireApproval: 'never',
    }),
  ],
});
What's happening:
  • We're creating an Agent instance with a name, model (gpt-5), and clear instructions about its purpose.
  • The agent's instructions tell it that it can access Braintrust and help with queries, inserts, updates, authentication, and fetching database information.
  • The tools array includes a hostedMcpTool that connects to the MCP server URL we created earlier.
  • The headers object includes the Composio API key for secure authentication with the MCP server.
  • requireApproval: 'never' means the agent can execute Braintrust operations without asking for permission each time, making interactions smoother.
9

Start chat loop and handle conversation

// Keep conversation state across turns
const conversationSession = new OpenAIConversationsSession();

// Simple CLI
const rl = readline.createInterface({
  input: process.stdin,
  output: process.stdout,
  prompt: 'You: ',
});

console.log('\nComposio Tool Router session created.');
console.log('\nChat started. Type your requests below.');
console.log("Commands: 'exit', 'quit', or 'q' to end\n");

try {
  const first = await run(agent, 'What can you help me with?', { session: conversationSession });
  console.log(`Assistant: ${first.finalOutput}\n`);
} catch (e) {
  console.error('Error:', e instanceof Error ? e.message : e, '\n');
}

rl.prompt();

rl.on('line', async (userInput) => {
  const text = userInput.trim();

  if (['exit', 'quit', 'q'].includes(text.toLowerCase())) {
    console.log('Goodbye!');
    rl.close();
    process.exit(0);
  }

  if (!text) {
    rl.prompt();
    return;
  }

  try {
    const result = await run(agent, text, { session: conversationSession });
    console.log(`\nAssistant: ${result.finalOutput}\n`);
  } catch (e) {
    console.error('Error:', e instanceof Error ? e.message : e, '\n');
  }

  rl.prompt();
});

rl.on('close', () => {
  console.log('\n👋 Session ended.');
  process.exit(0);
});
What's happening:
  • The program prints a session URL that you visit to authorize Braintrust.
  • After authorization, the chat begins.
  • Each message you type is processed by the agent using run().
  • The responses are printed to the console.
  • Typing exit, quit, or q cleanly ends the chat.

Complete Code

Here's the complete code to get you started with Braintrust and OpenAI Agents SDK:

import 'dotenv/config';
import { Composio } from '@composio/core';
import { OpenAIAgentsProvider } from '@composio/openai-agents';
import { Agent, hostedMcpTool, run, OpenAIConversationsSession } from '@openai/agents';
import * as readline from 'readline';

const composioApiKey = process.env.COMPOSIO_API_KEY;
const userId = process.env.USER_ID;

if (!composioApiKey) {
  throw new Error('COMPOSIO_API_KEY is not set. Create a .env file with COMPOSIO_API_KEY=your_key');
}
if (!userId) {
  throw new Error('USER_ID is not set');
}

// Initialize Composio
const composio = new Composio({
  apiKey: composioApiKey,
  provider: new OpenAIAgentsProvider(),
});

async function main() {
  // Create Tool Router session
  const session = await composio.create(userId as string, {
    toolkits: ['braintrust'],
  });
  const mcpUrl = session.mcp.url;

  // Configure agent with MCP tool
  const agent = new Agent({
    name: 'Assistant',
    model: 'gpt-5',
    instructions:
      'You are a helpful assistant that can access Braintrust. Help users perform Braintrust operations through natural language.',
    tools: [
      hostedMcpTool({
        serverLabel: 'tool_router',
        serverUrl: mcpUrl,
        headers: { 'x-api-key': composioApiKey },
        requireApproval: 'never',
      }),
    ],
  });

  // Keep conversation state across turns
  const conversationSession = new OpenAIConversationsSession();

  // Simple CLI
  const rl = readline.createInterface({
    input: process.stdin,
    output: process.stdout,
    prompt: 'You: ',
  });

  console.log('\nComposio Tool Router session created.');
  console.log('\nChat started. Type your requests below.');
  console.log("Commands: 'exit', 'quit', or 'q' to end\n");

  try {
    const first = await run(agent, 'What can you help me with?', { session: conversationSession });
    console.log(`Assistant: ${first.finalOutput}\n`);
  } catch (e) {
    console.error('Error:', e instanceof Error ? e.message : e, '\n');
  }

  rl.prompt();

  rl.on('line', async (userInput) => {
    const text = userInput.trim();

    if (['exit', 'quit', 'q'].includes(text.toLowerCase())) {
      console.log('Goodbye!');
      rl.close();
      process.exit(0);
    }

    if (!text) {
      rl.prompt();
      return;
    }

    try {
      const result = await run(agent, text, { session: conversationSession });
      console.log(`\nAssistant: ${result.finalOutput}\n`);
    } catch (e) {
      console.error('Error:', e instanceof Error ? e.message : e, '\n');
    }

    rl.prompt();
  });

  rl.on('close', () => {
    console.log('\nSession ended.');
    process.exit(0);
  });
}

main().catch((err) => {
  console.error('Fatal error:', err);
  process.exit(1);
});

Conclusion

This was a starter code for integrating Braintrust MCP with OpenAI Agents SDK to build a functional AI agent that can interact with Braintrust.

Key features:

  • Hosted MCP tool integration through Composio's Tool Router
  • SQLite session persistence for conversation history
  • Simple async chat loop for interactive testing
You can extend this by adding more toolkits, implementing custom business logic, or building a web interface around the agent.
TOOLS

Supported Tools

Every Braintrust action and event your agent gets out of the box.

Create Dataset

Create a named dataset under a Braintrust project, or return the existing dataset unchanged when its name already exists in that project.

Create Experiment

Create an experiment under a Braintrust project.

Create Project

Create a Braintrust project, or return the existing project unchanged when its name already exists.

Create Prompt

Upsert a versioned prompt configuration under a Braintrust project without invoking it.

Delete Project

Soft-delete a Braintrust project and return its deletion state.

Delete Experiment, Dataset, or Prompt

Delete one Braintrust experiment, dataset, or prompt by UUID for explicit lifecycle cleanup.

Fetch Events

Fetch one newest-to-oldest page of project log, experiment, or dataset events.

Get Braintrust Resource

Get one project, experiment, dataset, prompt, or function by UUID.

Insert Dataset Events

Upsert a batch of examples into an existing Braintrust dataset.

Insert Experiment Events

Upsert a batch of result events into an existing Braintrust experiment.

List Braintrust Resources

List and filter one page of projects, experiments, datasets, prompts, or functions visible to the connected Braintrust API key.

Query Braintrust Data

Run one bounded synchronous read-only BTQL query over Braintrust project logs, an experiment, or a dataset.

Summarize Experiment or Dataset

Return metadata and optional aggregate summaries for one Braintrust experiment or dataset.

Update Dataset

Update the name, description, tags, or metadata of a Braintrust dataset.

Update Experiment

Update selected metadata and linkage fields on a Braintrust experiment.

Update Project

Update selected fields of a Braintrust project.

Update Prompt

Update a Braintrust prompt's metadata or configuration without invoking it.

FAQ

Frequently asked questions

With a standalone Braintrust MCP server, the agents and LLMs can only access a fixed set of Braintrust tools tied to that server. However, with the Composio Tool Router, agents can dynamically load tools from Braintrust and many other apps based on the task at hand, all through a single MCP endpoint.

Yes, you can. OpenAI Agents SDK fully supports MCP integration. You get structured tool calling, message history handling, and model orchestration while Tool Router takes care of discovering and serving the right Braintrust tools.

Yes, absolutely. You can configure which Braintrust scopes and actions are allowed when connecting your account to Composio. You can also bring your own OAuth credentials or API configuration so you keep full control over what the agent can do.

All sensitive data such as tokens, keys, and configuration is fully encrypted at rest and in transit. Composio is SOC 2 Type 2 compliant and follows strict security practices so your Braintrust data and credentials are handled as safely as possible.

Start with Braintrust.It takes 30 seconds.

Managed auth, hosted MCP servers, and every Braintrust tool your agent needs.Free to start.

Start building