MCP Server

The tree-sitter-language-pack CLI includes an MCP server that exposes parsing, code intelligence extraction, language detection, and cache management as standard tools for AI agents. Use it to add code analysis to Claude, Cursor, VS Code, or any MCP-compatible application.

There are three ways to run it:

Bundled with the plugin — the coding-agent plugin registers the tree-sitter-language-pack MCP server for you and resolves the CLI automatically. Nothing to install by hand.
Direct MCP client config — point any MCP client at the published ts-pack CLI (see Installing the CLI below).
Hermes — for Hermes-based agents, install the runtime plugin with pip install tree-sitter-language-pack-hermes-plugin.

Installing the CLI

The ts-pack binary is published to every major registry. Install it with whichever fits your toolchain:

# Homebrew (macOS / Linux)
brew install xberg-io/tap/ts-pack

# Scoop (Windows) -- run these two in PowerShell
# scoop bucket add xberg https://github.com/xberg-io/scoop-bucket
# scoop install ts-pack

# npm (Node.js)
npm install -g @xberg-io/ts-pack-cli

# uv / uvx (Python)
uvx --from ts-pack-cli ts-pack --version

# Cargo (Rust)
cargo install ts-pack-cli

Once ts-pack is on your PATH, any of the client configs below will work.

What is MCP?

The Model Context Protocol (MCP) is an open standard for connecting AI applications to tools and data. The tree-sitter-language-pack MCP server provides tools for parsing source code, analyzing structure and symbols, and managing language packs — all through a unified interface.

Starting the Server

Stdio Transport (Default)

For local AI tools — Claude Desktop, Cursor, VS Code — use stdio transport:

ts-pack mcp --transport stdio

Stdio is the default, so ts-pack mcp is equivalent. The server runs as a subprocess and communicates over stdin/stdout with JSON-RPC messages. No network configuration needed.

HTTP Transport

For remote agents or team environments where stdio doesn’t work:

ts-pack mcp --transport http --host 127.0.0.1 --port 8011

The server listens on http://127.0.0.1:8011 (default). Change --host to 0.0.0.0 for network-wide access (use with caution).

Custom Configuration

Point the server at a language-pack.toml config file:

ts-pack mcp --config /path/to/language-pack.toml

This sets default languages and download preferences for all tool calls.

Registering with AI Tools

Claude Desktop / Claude Code

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "tree-sitter-language-pack": {
      "command": "ts-pack",
      "args": ["mcp", "--transport", "stdio"]
    }
  }
}

Or use the CLI to register automatically:

claude mcp add tree-sitter-language-pack -- ts-pack mcp --transport stdio

Restart Claude. The tree-sitter-language-pack tools appear in the Tools panel.

Cursor

Edit .cursor/mcp.json in your project root (or global Cursor settings):

{
  "mcpServers": {
    "tree-sitter-language-pack": {
      "command": "ts-pack",
      "args": ["mcp", "--transport", "stdio"]
    }
  }
}

Reload Cursor. Tools are now available in the AI chat.

VS Code / GitHub Copilot

Edit .vscode/settings.json or your VS Code global settings:

{
  "mcpServers": [
    {
      "name": "tree-sitter-language-pack",
      "command": "ts-pack",
      "args": ["mcp", "--transport", "stdio"]
    }
  ]
}

Then reference tools in GitHub Copilot chat or use the Tools panel.

Generic MCP Client

For a client that spawns the server over stdio, point it at the ts-pack binary:

{
  "mcpServers": [
    {
      "name": "tree-sitter-language-pack",
      "command": "ts-pack",
      "args": ["mcp", "--transport", "stdio"]
    }
  ]
}

For an HTTP client, start the server yourself (ts-pack mcp --transport http --port 8011) and connect to its URL:

{
  "mcpServers": [
    {
      "name": "tree-sitter-language-pack",
      "url": "http://127.0.0.1:8011"
    }
  ]
}

Tools

The MCP server exposes 8 tools for parsing, analysis, and management:

parse

Render the syntax tree as S-expression or JSON.

Parameter	Type	Description
`source`	string	Source code to parse
`language`	string	Language name (e.g., `python`, `rust`)
`format`	string	Output format: `sexp` or `json` (default: `sexp`)

process

Extract code intelligence: structure, imports, exports, symbols, docstrings, comments, diagnostics, and optionally chunk for LLMs.

Parameter	Type	Description
`source`	string	Source code to analyze
`language`	string	Language name
`all`	boolean	Enable every analysis feature; overrides the individual flags when true
`structure`	boolean	Extract structural items (default: true)
`imports`	boolean	Extract import statements (default: true)
`exports`	boolean	Extract export statements (default: true)
`comments`	boolean	Extract comments (default: false)
`symbols`	boolean	Extract symbol definitions (default: false)
`docstrings`	boolean	Extract docstrings (default: false)
`diagnostics`	boolean	Include parse diagnostics (default: false)
`data_extraction`	boolean	Hierarchical data extraction for data formats (default: false)
`chunk_max_size`	integer	Maximum chunk size in bytes; omit to disable chunking

There is no chunk_overlap parameter — chunks never overlap.

detect_language

Identify language from a file path, source content, or both. Both parameters are optional; supply at least one.

Parameter	Type	Description
`path`	string	File path or name, used for extension-based detection — optional
`content`	string	Source content, used for content-based detection — optional

list_languages

Enumerate available, downloaded, or manifest languages.

Parameter	Type	Description
`source`	string	Which set to query: `available` (default), `downloaded`, or `manifest`
`filter`	string	Substring filter applied to the result — optional

info

Get status of a specific language (known to this build, downloaded, cache directory).

Parameter	Type	Description
`language`	string	Language name (e.g., `python`)

download

Fetch language parsers for offline use.

Parameter	Type	Description
`languages`	array	List of language names — optional
`all`	boolean	Download every available language — optional
`groups`	array	Named groups to download. Only manifest-defined names are accepted; the published manifest defines exactly one, `all` — optional
`fresh`	boolean	Clean the cache before downloading (default: false)

cache_dir

Retrieve the local cache directory where parsers are stored. Takes no parameters.

Returns { "cache_dir": "<cache>/tree-sitter-language-pack/v{version}/libs" }.

clean_cache

Delete all cached parsers. Takes no parameters — there is no per-language variant.

Returns { "cache_dir": "…", "status": "cleared" }.

Resources

The MCP server provides read-only resources for browsing the language catalog:

ts-pack://languages — list of all 371 available languages with extensions and aliases
ts-pack://languages/downloaded — list of user-downloaded languages
ts-pack://language/{name} — status of a specific language (template resource)

Prompt

The MCP server includes a built-in prompt template:

analyze-code

Returns a user message instructing the agent to call the process tool with all=true, then summarize the design, key entry points, and any issues. The prompt does not take the source code itself — the agent supplies it when it calls process.

Parameter	Type	Required	Description
`language`	string	Yes	Language name. Supports argument completion against the language list.
`focus`	string	No	Free-form area to emphasize, e.g. `security` or `public API`.

focus is not a closed enum — whatever string you pass is appended to the prompt as “Pay particular attention to: …”. Omitting language falls back to the literal phrase “the file’s language”.

Using the Plugin Instead

For most users, installing the tree-sitter-language-pack plugin from the self-hosted xberg-io/tree-sitter-language-pack marketplace is simpler than manual MCP registration. The plugin ships a launcher script (scripts/mcp-launch.sh) that resolves the ts-pack CLI at runtime — via a cached binary, npx @xberg-io/ts-pack-cli, uvx --from ts-pack-cli ts-pack, Homebrew, or a prebuilt release download — and registers the tree-sitter-language-pack server for you.

See AI Coding Assistants for installation steps.

Next Steps

CLI Guide — full CLI command reference
Parsing Code — understanding syntax trees
Code Intelligence — extract structure and symbols
Installation — install ts-pack-cli