Skip to main content

BuildBuddy MCP server

BuildBuddy's MCP server provides a suite of tools that enable agents to explore build and test metadata, logs, artifacts, target statuses, and more.

MCP enables powerful agentic workflows without requiring agents to hand-write BuildBuddy API requests. For example, a local coding agent can use BuildBuddy's MCP tools to check failing CI tests on BuildBuddy and then fix issues automatically, or drill into performance issues. Coding agents can also be triggered as part of CI, using BuildBuddy's observability tools to quickly produce actionable diagnostics and insights.

Available MCP tools​

The MCP server exposes tools for inspecting build metadata, test logs, artifacts, and more, from a single BuildBuddy org.

Most BuildBuddy APIs are also available as MCP tools. For the underlying API reference or field-level details, see the Enterprise API docs.

The available tools are subject to change. It's strongly recommended not to write agent instructions (AGENTS.md, skills, etc.) referencing specific MCP tool names. Instead, give agents generic instructions and let them explore build data using the currently available tools.

Endpoint​

Cloud BuildBuddy MCP servers are available at:

https://<org>.buildbuddy.io/mcp

Replace <org> with your organization's BuildBuddy slug.

If you're running BuildBuddy on-prem, use your own BuildBuddy base URL instead:

https://<your-buildbuddy-host>/mcp

Authentication​

Authenticate with a BuildBuddy API key.

Most MCP clients are easiest to configure with this header:

Authorization: Bearer <api-key>

You can create API keys on your organization settings page. For more details, see the Authentication Guide.

Auth best practices​

If you want the strongest guarantee that a BuildBuddy API key is not exposed to an agent, run an MCP relay server that injects the Authorization: Bearer <api-key> header on outbound requests, then point the agent at that relay. Ideally, run the relay in an isolated environment that the agent cannot inspect directly.

For the API key, a good pattern is to use a dedicated BuildBuddy user with the Reader role and a user-owned API key, if your org has user-owned keys enabled. This ensures that the narrowest possible scope for the key.

Configuration​

See the relevant MCP configuration instructions below for your agent.

Many other coding agents support MCP as well - prompting the agent itself for MCP setup will often yield usable instructions or references.

Self-hosted setup​

For self-hosted BuildBuddy, enable both the public API and MCP endpoint:

config.yaml
api:
  enable_api: true
  enable_mcp: true

Then point your MCP client at:

https://<your-buildbuddy-host>/mcp