Code Tools MCP Server

• Local-only MCP server exposing core coding tools for LLMs via STDIO.
• Tools: list_directory, read_file, write_file, grep, ripgrep, glob, edit, read_many_files.

Codex CLI on windows sort of sucks because it relies on writing PS / Python scripts for basic read, write, grep operations. This MCP server exposes those standard tools making Codex CLI faster on Windows. You can use it on Linux or Mac, it will work but may not be necessary.

This is without warranty, any issues or bugs should be reported to the repository but be aware of the risks and use it at your own risk.

Install

• Global: npm i -g code-tools-mcp
• One-off: npx code-tools-mcp

Run

• code-tools-mcp --root C:/path/to/workspace

CODEX CLI Config Example

[mcp_servers.code-tools]
command = "{path to npm.cmd}"
args = [ "-y", "code-tools-mcp"]
env = { APPDATA = "C:\\Users\\{username}\\AppData\\Roaming", LOCALAPPDATA = "C:\\Users\\{username}\\AppData\\Local", HOME = "C:\\Users\\{username}", SystemRoot = "C:\\Windows", ComSpec = "C:\\Windows\\System32\\cmd.exe" }
startup_timeout_ms = 20_000

Workspace root is auto-detected:

• If CODE_TOOLS_MCP_ROOT is set, it wins.
• Else if a CLI flag is passed, it’s used: --root C:/path/to/workspace (or -r).
• Else, the server looks upward from the current working directory for a .git folder and uses that directory as root.
• Else, it defaults to the current working directory.

Claude config example without env var (pass a root flag):

{
  "mcpServers": {
    "code-tools": {
      "command": "node",
      "args": [
        "C:/Users/adity/Projects/code-tools-mcp/dist/index.js",
        "--root",
        "C:/Users/adity/Projects/code-tools-mcp"
      ]
    }
  }
}

Claude Desktop Config Example

Add to your Claude config JSON:

{
  "mcpServers": {
    "code-tools": {
      "command": "node",
      "args": ["/ABSOLUTE/PATH/code-tools-mcp/dist/index.js"],
      "env": { "CODE_TOOLS_MCP_ROOT": "/ABSOLUTE/PATH/TO/YOUR/WORKSPACE" }
    }
  }
}

Notes

• Uses STDIO transport; avoid console.log (stdout). Any diagnostics are written to stderr.
• File operations are restricted to the workspace root.
• read_file caps file size at 2MB; search ignores common large folders.
• Only .gitignore patterns are respected (.geminiignore is NOT supported).

---

TODO:

• Repomap using AST

---

Tools

list_directory

Lists directory contents with directories first, respects .gitignore.

Parameters:

• path (string, required): Absolute path to directory
• ignore (string[], optional): Glob patterns to ignore
• file_filtering_options (object, optional): { respect_git_ignore?: boolean }

Example:

await client.callTool('list_directory', {
  path: '/path/to/workspace/src'
});

read_file

Reads a file with optional pagination. Binary-aware.

Parameters:

• absolute_path (string, required): Absolute path to file
• offset (number, optional): Starting line (0-based)
• limit (number, optional): Number of lines to return (max 2000)
• allow_ignored (boolean, optional): Allow reading .gitignore-ignored files (default: false)

Features:

• Returns nextOffset in structured content for paginated reads
• Includes summary with line count in all responses

Example:

await client.callTool('read_file', {
  absolute_path: '/path/to/workspace/src/index.ts',
  offset: 0,
  limit: 100
});

write_file

Creates or overwrites a file with preview mode.

Parameters:

• file_path (string, required): Absolute path of file to write
• content (string, required): Full file content
• apply (boolean, default: false): If false, returns diff preview without writing
• overwrite (boolean, default: true): Allow overwriting existing files

Features:

• Preview mode by default (set apply: true to write)
• Returns diff in structured content

Example:

await client.callTool('write_file', {
  file_path: '/path/to/workspace/src/new-file.ts',
  content: 'export const foo = "bar";',
  apply: true
});

grep

Text/regex search with smart-case support.

Parameters:

• pattern (string, required): Search pattern (plain text or regex if regex: true)
• path (string, optional): Directory path relative to workspace
• include (string, optional): Glob filter (e.g., **/*.{ts,tsx})
• exclude (string | string[], optional): Glob exclusion patterns
• regex (boolean, default: false): Treat pattern as regex
• ignore_case (boolean, optional): Case-insensitive search. If undefined, uses smart-case (case-sensitive if pattern has uppercase)
• max_matches (number, default: 2000): Maximum matches to return

Features:

• Smart-case: automatically case-sensitive if pattern contains uppercase letters
• Skips binary files
• Respects .gitignore
• Returns truncated: true if max_matches reached

Example:

await client.callTool('grep', {
  pattern: 'function.*async',
  regex: true,
  include: '**/*.ts',
  max_matches: 100
});

ripgrep

Fast regex search using ripgrep binary (falls back to grep if unavailable).

Parameters:

• pattern (string, required): Regular expression
• path (string, optional): Directory to search
• include (string | string[], optional): Glob include patterns
• exclude (string | string[], optional): Glob exclude patterns
• ignore_case (boolean, optional): If undefined, uses --smart-case
• max_matches (number, default: 20000): Maximum matches to return

Features:

• Uses --smart-case by default when ignore_case not specified
• Returns truncated: true if max_matches reached
• Falls back to JavaScript grep if ripgrep unavailable
• Captures stderr in structuredContent.stderr

Example:

await client.callTool('ripgrep', {
  pattern: 'TODO|FIXME',
  include: ['**/*.ts', '**/*.tsx'],
  exclude: '**/node_modules/**'
});

glob

Matches files by glob pattern with two-tier sorting.

Parameters:

• pattern (string, required): Glob pattern (e.g., **/*.ts)
• path (string, optional): Directory to search (defaults to workspace root)
• case_sensitive (boolean, default: false): Case-sensitive matching
• respect_git_ignore (boolean, default: true): Respect .gitignore

Features:

• Two-tier sorting: recent files (< 24h) newest-first, then older files alphabetically
• Reports gitIgnoredCount when no files found

Example:

await client.callTool('glob', {
  pattern: '**/*.{ts,tsx}',
  path: '/path/to/workspace/src'
});

edit

Targeted text replacement with preview mode.

Parameters:

• file_path (string, required): Absolute path to file
• old_string (string, required): Text to replace (empty string creates new file)
• new_string (string, required): Replacement text
• expected_replacements (number, default: 1): Expected number of replacements
• apply (boolean, default: false): If false, returns diff preview

Features:

• Preview mode by default
• Detects no-change edits
• Returns occurrences in structured content

Example:

await client.callTool('edit', {
  file_path: '/path/to/workspace/src/index.ts',
  old_string: 'const foo = "old";',
  new_string: 'const foo = "new";',
  apply: true
});

read_many_files

Reads multiple files by glob patterns, concatenated output.

Parameters:

• paths (string[], required): Array of file/directory globs
• include (string[], optional): Additional glob patterns to include
• exclude (string[], optional): Glob patterns to exclude
• useDefaultExcludes (boolean, default: true): Apply default excludes (node_modules, dist, .git, etc.)
• file_filtering_options (object, optional): { respect_git_ignore?: boolean }

Features:

• 2MB total output cap
• Workspace containment re-check for safety
• Aggregated skip reasons in structured content (skipCounts)
• Returns truncated: true when total cap reached

Example:

await client.callTool('read_many_files', {
  paths: ['src/**/*.ts'],
  exclude: ['**/*.test.ts']
});

---

Using These Tools from an MCP Client

Tool discovery: clients call tools/list (supports cursor pagination) and may receive tools/list_changed when the set changes.

TypeScript Example

import { Client } from '@modelcontextprotocol/sdk/client/index.js';
import { StdioClientTransport } from '@modelcontextprotocol/sdk/client/stdio.js';

// Connect to the server
const transport = new StdioClientTransport({
  command: 'node',
  args: ['/path/to/code-tools-mcp/dist/index.js'],
  env: { CODE_TOOLS_MCP_ROOT: '/path/to/workspace' }
});

const client = new Client({
  name: 'example-client',
  version: '1.0.0'
}, {
  capabilities: {}
});

await client.connect(transport);

// List available tools
const tools = await client.listTools();
console.log('Available tools:', tools.tools.map(t => t.name));

// Call a tool
const result = await client.callTool({
  name: 'read_file',
  arguments: {
    absolute_path: '/path/to/workspace/src/index.ts'
  }
});

console.log('Result:', result);

Pagination Example

// Read large file in chunks
let offset = 0;
const limit = 100;

while (true) {
  const result = await client.callTool({
    name: 'read_file',
    arguments: {
      absolute_path: '/path/to/large-file.ts',
      offset,
      limit
    }
  });

  console.log(result.content[0].text);

  // Check if there's more to read
  const nextOffset = result.structuredContent?.nextOffset;
  if (!nextOffset) break;

  offset = nextOffset;
}

---

Tool Stability

• Tool names are stable and will not change in minor versions
• Parameter names are stable - new optional parameters may be added
• Descriptions are concise - detailed parameter docs are in JSON Schema (via Zod .describe())
• Structured content includes summary field for all tools where applicable
• Defaults and limits are documented in parameter descriptions:
  • read_file: 2MB file cap, 2000 line limit per request
  • read_many_files: 2MB total output cap, 1MB per file
  • grep: 2000 matches default, configurable via max_matches
  • ripgrep: 20000 matches default, configurable via max_matches

---

TypeScript Types

All tool input and output types are available in src/types/tools.ts:

import type {
  ReadFileInput,
  ReadFileStructuredOutput,
  GrepInput,
  GrepStructuredOutput,
  // ... other types
} from 'code-tools-mcp/types/tools';