Braintrust MCP
AI model by Braintrust
Connect your coding agent to your AI stack.
Braintrust MCP is a server that connects coding agents and IDEs to the Braintrust AI stack, enabling developers to query logs and run evals directly from their local environment. It is ideal for teams using agents like Claude Code to accelerate production debugging and prompt optimization.
Model specs
The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.
Category: Developer Tools
- Sizes and variants
- Braintrust MCP server
- Access
- IDE integration, MCP server, coding agents
- Supported Agents
- Claude Code, Codex, OpenCode, Pi, Antigravity, Grok
- Primary Interface
- Model Context Protocol (MCP)
- Core Functions
- Log querying, eval execution, prompt updates
- Integration Method
- bt tracing integration
- Query logs and run evals directly from the IDE
- Connects coding agents to the Braintrust AI stack
- Supports Claude Code, Codex, and other MCP-compatible tools
- Enables prompt updates from local development environments
Overview
Braintrust MCP is a Model Context Protocol server designed to bridge the gap between production observability and local development. It allows engineers and businesses to connect their coding agents and IDEs directly to the Braintrust AI stack to query production logs, run evaluations, and update prompts without switching contexts. By providing agents with access to project data and investigation skills, it facilitates faster debugging and optimization of AI applications. The server is compatible with popular tools such as Claude Code and Codex, enabling a tighter feedback loop for fixing production failures and testing improvements against real-world data.
Developer Workflow Integration
- Direct Log Querying: The MCP server allows developers to query production logs and traces directly from their IDE or coding agent.
- In-IDE Evals: Engineers can execute evaluations and experiments against Braintrust datasets without leaving their local development environment.
- Prompt Management: The server enables the updating of prompts and model configurations directly from the developer's coding interface.
- Agent Connectivity: MCP connects coding agents to Braintrust's built-in investigation skills and project context.
Compatibility and Tools
- Claude Code Support: Braintrust MCP is compatible with Claude Code for integrated agent-led investigations.
- Codex Integration: The server integrates with Codex to combine repository data with production observability.
- Protocol Compatibility: Developers can use the MCP server with any tool that supports the Model Context Protocol.
- Project Tool Access: The server provides access to project-specific tools that can read and write Braintrust state.
Use cases
- Developers query production logs and traces directly from their IDE to debug agent failures.
- Engineers run evaluations against real datasets to test code changes before deployment.
- AI teams update and version prompts within their local workflow using production context.
- Coding agents use Braintrust investigation skills to identify the root cause of production regressions.
Specifications
- Sizes and variants
- Braintrust MCP server
- Access
- IDE integration, MCP server, coding agents
- Supported Agents
- Claude Code, Codex, OpenCode, Pi, Antigravity, Grok
- Primary Interface
- Model Context Protocol (MCP)
- Core Functions
- Log querying, eval execution, prompt updates
- Integration Method
- bt tracing integration
How it works
- Project Setup: Set up a dedicated project for coding agent sessions and enable tracing from the development stack.
- Agent Connection: Connect the coding agent (such as Claude Code or Codex) through the Braintrust MCP server.
- Investigate and Fix: Use the agent to query production logs, investigate failures, and test fixes against real cases.
Pricing
- Starter
- $0 / month
- Pro
- $249 / month
- Enterprise
- Custom pricing
FAQ
What is Braintrust MCP?
Braintrust MCP is a server that connects your coding agent and IDE to your Braintrust AI stack, allowing you to query logs, run evals, and update prompts directly.
Which coding agents work with Braintrust MCP?
It is compatible with several coding agents including Claude Code, Codex, OpenCode, Pi, Antigravity, and Grok through the bt tracing integration.
How does Braintrust MCP differ from the Loop agent?
Loop starts with product context and investigation skills already loaded, while MCP + Codex combines project data with repository, terminal, and connected tools.
Added by AntSpace from public sources (braintrust.dev).