Nx MCP

Nx MCP Server

A Model Context Protocol server implementation for Nx.

Overview

The Nx MCP server gives LLMs deep access to your monorepo’s structure: project relationships, file mappings, runnable tasks, ownership info, tech stacks, Nx generators, and even Nx documentation. With this context, LLMs can generate code tailored to your stack, understand the impact of a change, and apply modifications across connected files with precision. This is possible because Nx already understands the higher-level architecture of your workspace, and monorepos bring all relevant projects into one place.

Read more in our blog post and in our docs.

Installation and Usage

There are two ways to use this MCP server:

a) Run it via the nx-mcp package

Simply invoke the MCP server via npx or your package manager's equivalent.

Here's an example of a mcp.json configuration:

{

"servers": { "nx-mcp": { "type": "stdio", "command": "npx", "args": ["nx-mcp@latest"] } } }

Claude Code

claude mcp add nx-mcp npx nx-mcp@latest

VSCode

code --add-mcp '{"name":"nx-mcp","command":"npx","args":["nx-mcp"]}'

Warp

Go to Settings -> AI -> Manage MCP Servers -> + Add to add an MCP Server.

Alternatively, use the slash command /add-mcp in the Warp Agent prompt.

{

"nx-mcp": { "command": "npx", "args": ["nx-mcp@latest"] } }

Refer to your AI tool's documentation for how to register an MCP server. For example, Cursor or Claude Desktop support MCP.

If you want to host the server instead of communicating via stdio, use --transport http and optionally --port (default 9921). Clients connect to http://localhost: /mcp using the Streamable HTTP transport. It supports multiple concurrent connections, allowing different clients to connect simultaneously with independent sessions. The deprecated SSE transport (--sse, --transport sse) has been removed.

Run nx-mcp --help to see what options are available.

b) Use the Nx Console extension

If you're using Cursor you can directly install the Nx Console extension which automatically manages the MCP server for you.

More info:

• Install Nx Console
• Configure Cursor to use the nx-mcp

Environment Files

By default, Nx MCP loads .local.env, .env.local, and .env from the workspace root. To skip this loading and preserve the environment supplied by your launcher, set NX_LOAD_DOT_ENV_FILES to the string "false" before starting the server:

{

"servers": { "nx-mcp": { "type": "stdio", "command": "npx", "args": ["nx-mcp@latest"], "env": { "NX_LOAD_DOT_ENV_FILES": "false" } } } }

This also skips root dotenv loading when Nx Console resolves its IDE socket path. If you use a custom socket directory, supply NX_SOCKET_DIR explicitly to both the server and the editor process. The setting does not undo environment variables loaded by a parent process or by other tools.

Minimal Mode (Default)

By default, the Nx MCP server runs in minimal mode, which hides workspace analysis tools that overlap with AI tool skills/instructions. This reduces tool clutter and context window usage for AI agents that already have access to equivalent functionality through skills.

The following tools are hidden in minimal mode:

• nx_available_plugins
• nx_workspace_path
• nx_workspace
• nx_project_details
• nx_generators
• nx_generator_schema

To disable minimal mode and expose all tools, use the --no-minimal flag:

npx nx-mcp@latest --no-minimal

Available Tools

The Nx MCP server provides a comprehensive set of tools for interacting with your Nx workspace.

Always Available

• nx_docs: Returns documentation sections relevant to user queries about Nx
• nx_current_running_tasks_details: Lists currently running Nx TUI processes and task statuses
• nx_current_running_task_output: Returns terminal output for specific running tasks
• nx_visualize_graph: Visualizes the Nx graph (requires running IDE instance)

Hidden in Minimal Mode (use --no-minimal to enable)

• nx_available_plugins: Lists available Nx plugins from the core team and local workspace plugins
• nx_workspace_path: Returns the path to the Nx workspace root
• nx_workspace: Returns readable representation of project graph and nx.json configuration
• nx_project_details: Returns complete project configuration in JSON format for a given project
• nx_generators: Returns list of generators relevant to user queries
• nx_generator_schema: Returns detailed JSON schema for a specific Nx generator

Nx Cloud Tools (only available w/ Nx Cloud enabled)

These tools provide insights and interactions with your Nx Cloud CI/CD data:

• ci_information: Retrieves CI pipeline execution information for the current branch or a specific Nx Cloud URL. Returns failed tasks with runId metadata and supports a select parameter for field selection with pagination.
• ci_task_output: Retrieves terminal output for a specific failed CI task, using a runId from ci_information or resolving it from the branch / URL context.
• update_self_healing_fix: Apply, reject, or request a rerun for a self-healing CI fix from Nx Cloud

When no workspace path is specified, only the nx_docs and nx_available_plugins tools will be available.

Available Resources

When connected to an Nx Cloud-enabled workspace, the Nx MCP server automatically exposes recent CI Pipeline Executions (CIPEs) as MCP resources. Resources appear in your AI tool's resource picker, allowing the LLM to access detailed information about CI runs including failed tasks, terminal output, and affected files.

Contributing & Development

Contributions are welcome! Please see the Nx Console contribution guide for more details.

The basic steps are:

1. Clone the Nx Console repository and follow installation steps
2. Build the nx-mcp using nx run nx-mcp:build (or nx run nx-mcp:build:debug for debugging with source maps)
3. Use the MCP Inspector to test out your changes

Configuration File

You can also configure the MCP server via a JSON file at .nx/nx-mcp-config.json in your workspace root. This is useful when the MCP server binary is managed by a plugin or extension where you can't pass CLI args directly.

{

"tools": ["nx_docs", "!cloud_*"], "minimal": true, "debugLogs": false, "disableTelemetry": true, "transport": "http", "port": 9922 }

All keys are optional and match the CLI option names. CLI arguments take precedence over config file values.

License

MIT