Skip to content

AI Agent Access (MCP)

Don't use MCP as an API

The EngFlow MCP endpoint is designed exclusively for AI agent access. Do not build scripts, automations, or other integrations against it. The available tools and their interfaces may change without notice.

What is MCP?

The Model Context Protocol (MCP) is an open standard for connecting AI agents to external data sources and tools. It allows agents to query your EngFlow cluster directly as part of their workflow. For a full protocol specification, see the MCP documentation.

What can AI agents do with EngFlow?

When connected to your EngFlow cluster via MCP, AI agents can help you investigate and resolve development issues using the features enabled on your cluster:

Debug build and test failures: Agents can examine build results, read invocation logs, and inspect target details to help you understand why a build or test failed and then fix the code locally, all in one workflow.

Identify and debug flaky targets: Agents can look at historical test results across invocations to identify intermittently failing targets, surface failure patterns, and suggest next steps for investigation.

  • Example: "Which targets have been flaky this week? What's causing the failures?"
  • Example: "Is the //src/lib:parser_test flaky on my branch or has it always been flaky?"

Diagnose build speed bottlenecks: Agents can analyze build data to pinpoint where time is being spent and suggest optimizations.

  • Example: "Why is my build slow? What are the longest-running actions?"

How agents access data

The EngFlow MCP server provides programmatic access to the same data visible in the Build and Test UI. A useful rule of thumb: if you can see it in the UI, your agent can likely access it via MCP.

The specific tools available to your agent depend on the features enabled on your cluster. When in doubt, you may ask your agent to describe what tools are available to it.

To enable more tools, talk to your cluster administrator.

Security

The EngFlow MCP server operates with the same permissions as the authenticated user. This means any AI agent connected via MCP can access the same build results, logs, and test data that you can see in the Build and Test UI.

Danger

Note that your AI agent may have access to sensitive information such as source files, environment variables, command-line arguments, and build and test output. Ensure that any AI agent you connect to your cluster is one you trust with access to this data.

Prerequisites

To use the EngFlow MCP server, your cluster must meet the following requirements:

  • Authentication: Your cluster must either have no HTTP authentication configured, or must have JWT token generation enabled.
  • Feature Enabled: MCP must be enabled on your cluster. Ask your cluster administrator or reach out to EngFlow support to enable it.

Setup

MCP setup is cluster-specific and takes only a few minutes. Your cluster's Getting Started page provides copy-paste configuration for supported AI clients including Claude Code, Cursor, and Windsurf.

To set up MCP access, visit your cluster's /getting-started#mcp page.