Skip to main content
Glama

Bazel MCP Server

A local MCP server that exposes functionality of the Bazel build system to MCP-enabled AI agents.

This is helpful when MCP environments either don't have an existing command-line tool, or where the invoked shell has a misconfigured environment that prevents Bazel from being used.

Tools

The Bazel MCP Server provides the following tools:

  • bazel_build_target: Build specified Bazel targets

  • bazel_query_target: Query the dependency graph for targets matching a pattern

  • bazel_test_target: Run tests for specified targets

  • bazel_list_targets: List all available targets in the workspace (requires path parameter, use "//" for all targets)

  • bazel_fetch_dependencies: Fetch external dependencies

  • bazel_set_workspace_path: Change the Bazel workspace path at runtime

Each command (except bazel_set_workspace_path) supports an optional additionalArgs parameter that allows passing additional arguments to the underlying Bazel command. This is useful for specifying flags like --verbose_failures or --test_output=all.

Related MCP server: bazel-mcp

Usage

Installation

Using with Cursor

Add the following to .cursor/mcp.json.

You don't need to provide the workspace path, as the LLM can use set_workspace_path to change the workspace path at runtime.

The bazel binary usually gets picked up automatically, but if you run into issues, you can provide the path to the bazel binary using the --bazel_path flag.

:warning: Note that this will not work when using Cursor with Remote SSH sessions, since it runs the MCP server locally.

{
  "mcpServers": {
    "bazel": {
      "command": "npx",
      "args": [
        "-y",
        "github:nacgarg/bazel-mcp-server",

        // If you need to specify the bazel binary path
        "--bazel_path", 
        "/absolute/path/to/your/bazel/binary",

        // If you need to specify the workspace path
        "--workspace_path",
        "/absolute/path/to/your/bazel/workspace"

        // See Configuration Table below for more options
      ]
    }
  }
}

Using with Claude Desktop

You can use the same configuration as above with Claude Desktop.

Launching standalone

# Run directly from GitHub (no installation needed)
npx -y github:nacgarg/bazel-mcp-server

# From source
git clone https://github.com/nacgarg/bazel-mcp-server.git
cd bazel-mcp-server
npm install
npm run build
dist/index.js

Configuration

This MCP server allows a couple different configuration methods. They will be used in the following order:

  1. Command line arguments

  2. Environment variables

  3. Configuration file

Configuration Table

CLI Argument

Environment Variable

Configuration File Key

Description

--bazel_path

MCP_BAZEL_PATH

bazel_path

The path to the Bazel binary to use.

--workspace_path

MCP_WORKSPACE_PATH

workspace_path

The path to the Bazel workspace to use.

--workspace_config

MCP_WORKSPACE_CONFIG

workspace_config

The configuration of the workspace to use. By default, this uses the .bazelrc file in the workspace root.

--log_path

MCP_LOG_PATH

log_path

The path to write server logs to.

Debugging

Set the DEBUG=true environment variable to enable verbose logging to the console.

Setting the log path is also helpful for debugging with clients that don't print logs to the console (looking at you, Cursor).

License

This project is licensed under the MIT License. See the LICENSE file for details.

Available Tools

6 tools
bazel_build_targetC

Build specified Bazel targets

Inputs: targets (required), additionalArgs.

bazel_fetch_dependenciesB

Fetch Bazel external dependencies

Inputs: targets, additionalArgs.

bazel_list_targetsB

List all available Bazel targets under a given path

Inputs: path (required), additionalArgs.

bazel_query_targetA

Query the Bazel dependency graph for targets matching a pattern

Inputs: pattern (required), additionalArgs.

bazel_set_workspace_pathB

Set the current Bazel workspace path for subsequent commands

Inputs: path (required).

bazel_test_targetC

Run Bazel tests for specified targets

Inputs: targets (required), additionalArgs.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.6/5.0

    Scored across 6 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose with no ambiguity: building, fetching dependencies, listing targets, querying the dependency graph, setting workspace path, and testing. The descriptions clearly differentiate their functions, making misselection unlikely.

    Naming Consistency5/5

    All tools follow a consistent 'bazel_verb_noun' pattern (e.g., bazel_build_target, bazel_fetch_dependencies), using snake_case throughout. This predictable naming makes the tool set easy to navigate and understand.

    Tool Count5/5

    With 6 tools, this server is well-scoped for Bazel operations, covering core workflows like building, testing, querying, and workspace management. Each tool earns its place without feeling excessive or insufficient for the domain.

    Completeness4/5

    The tool set provides strong coverage for essential Bazel tasks, including build, test, query, and dependency management. A minor gap exists in operations like cleaning or caching, but agents can work around this with the available tools for most common workflows.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      B
      quality
      D
      maintenance
      A local-first MCP server that provides AI agents with safe codebase access through file discovery, hybrid lexical-semantic search, and project introspection. It features durable local memory and semantic indexing while keeping all data and processing entirely on your local machine.
      74
      27 npm
      6
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      A lightweight build daemon that exposes project build flows as MCP tools, enabling AI assistants to drive builds, check status, and react to results.
      MIT