MCP Claude Code

Claude Code-style coding agent over MCP: read/edit files, run commands, search code, delegate to sub-agents.

Unverified

stdio (local)

No auth

Stale

Python

View repo 301

Add to your client

Copy the config for your MCP client and paste it into its config file.

Install / run

uvx --from mcp-claude-code claudecode --install

Paste into ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "mcp-claude-code": {
      "command": "uvx",
      "args": [
        "--from",
        "mcp-claude-code",
        "claudecode",
        "--allow-path",
        "/path/allow",
        "--project",
        "/path/project1",
        "--project",
        "/path/project2"
      ]
    }
  }
}

Requires `uv` (the Python package runner). Install it from https://docs.astral.sh/uv/ if `uvx` is not found.

Step-by-step guides: Add to Claude Desktop · Add to Cursor · Add to Windsurf

Before you start

Python 3.12 or later
Claude Desktop application (or another MCP client)
uv/uvx or pip to install the mcp-claude-code package
Optional: ripgrep for enhanced search performance
Optional: an LLM provider API key (LiteLLM-compatible) to enable the agent tool

About MCP Claude Code

An MCP server implementing Claude Code-like functionality so that MCP clients can analyze codebases, make targeted file edits with permission handling, run shell commands, search code with ripgrep and AST context, read/edit Jupyter notebooks, manage structured task lists, and delegate complex tasks to concurrent read-only sub-agents using any LiteLLM-compatible model. Filesystem access is restricted to directories specified via --allow-path.

Tools & capabilities (16)

read

Read file contents with line numbers, offset, and limit capabilities.

write

Create or overwrite files.

edit

Make line-based edits to text files.

multi_edit

Make multiple precise text replacements in a single file operation with atomic transactions.

directory_tree

Get a recursive tree view of directories.

grep

Fast pattern search in files with ripgrep integration for best performance.

content_replace

Replace patterns in file contents.

grep_ast

Search code with AST context showing matches within functions, classes, and other structures.

run_command

Execute shell commands (also used for directory creation, file moving, and directory listing).

notebook_read

Extract and read source code from all cells in a Jupyter notebook with outputs.

notebook_edit

Edit, insert, or delete cells in a Jupyter notebook.

think

Structured space for complex reasoning and analysis without making changes.

dispatch_agent

Launch one or more agents that can perform tasks using read-only tools concurrently.

batch

Execute multiple tool invocations in parallel or serially in a single request.

todo_write

Create and manage a structured task list.

todo_read

Read a structured task list.

When to use it

Give Claude Desktop the ability to read, edit, and refactor files in a local project with permission controls.
Search large codebases quickly using ripgrep and AST-aware matching.
Run build, test, or shell commands and manage files from within an MCP client.
Read and edit Jupyter notebooks, including cell outputs.
Delegate complex multi-step research tasks to concurrent sub-agents backed by a configurable LLM.

Security notes

Access is restricted to directories passed via --allow-path; file modifications and command execution trigger permission handling, with input validation and sanitization. The run_command tool executes arbitrary shell commands within allowed paths, so only grant access to directories you trust. The optional agent tool requires an LLM provider API key (passed via --agent-api-key or environment variables such as OPENAI_API_KEY / ANTHROPIC_API_KEY).

MCP Claude Code FAQ

How do I install and configure it?

The recommended method is uvx: run `uvx --from mcp-claude-code claudecode --install` to auto-configure Claude Desktop, or add a manual `mcpServers` entry with command `uvx` and args `--from mcp-claude-code claudecode --allow-path /path/allow`. You can also `pip install mcp-claude-code` and use the `claudecode` command directly.

How is filesystem access secured?

The server only accesses directories you explicitly pass via `--allow-path` (multiple allowed). File modifications and command execution use permission handling, and inputs are validated and sanitized.

How do I enable the agent (sub-agent) tool?

Add `--enable-agent-tool` and configure an LLM provider with `--agent-model` (LiteLLM format, e.g. `openai/gpt-4o` or `anthropic/claude-3-5-sonnet-20241022`) plus `--agent-api-key`, or use environment variables like OPENAI_API_KEY / ANTHROPIC_API_KEY. It is disabled by default.

What transports are supported?

The default transport is stdio; the `--transport` option also supports `sse`.

#claude-code #coding-agent #file-operations #code-search #ripgrep #shell #jupyter #agent-delegation #litellm

Alternatives to MCP Claude Code

Compare all alternatives →

Playwright MCP Server

Developer Tools

24k

Microsoft's official browser-automation MCP using Playwright's accessibility tree (no vision model).

Featured

Verified

stdio (local)

No auth

TypeScript

12 tools

Updated 13 days agoRepo

Context7 MCP Server

Developer Tools

28k

Up-to-date, version-specific library documentation injected into your coding agent.

Verified

stdio (local)

API key

TypeScript

2 tools

Updated 17 days agoRepo

Serena

Developer Tools

12k

LSP-powered coding agent toolkit: semantic symbol search, references and structural edits.

Verified

stdio (local)

No auth

Python

11 tools

Updated 15 days agoRepo

Compare MCP Claude Code with:

vs Playwright MCP Server vs Context7 MCP Server vs Serena vs Mobile MCP