Skip to content
Back to search

mcpb

Desktop Extensions: One-click local MCP server installation in desktop apps

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/modelcontextprotocol--mcpb/badge.svg)](/listings/modelcontextprotocol--mcpb)

HTML

<a href="/listings/modelcontextprotocol--mcpb"><img src="/api/listings/modelcontextprotocol--mcpb/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

# Clone the repository
git clone https://github.com/anthropics/mcpb.git
cd mcpb

# Install dependencies
yarn

# Build the project
yarn build

# Run tests
yarn test

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 172 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # MCP Bundles (MCPB)
  2. > **⚠️ IMPORTANT NOTICE: This project is being renamed from DXT (Desktop Extensions) to MCPB (MCP Bundles)**
  3. >
  4. > If you're looking for the DXT tools, they have been renamed to MCPB. Please update your dependencies and tooling:
  5. >
  6. > - `dxt` CLI is now `mcpb`
  7. > - `.dxt` files are now `.mcpb` files
  8. > - `@anthropic-ai/dxt` package will be moved to `@anthropic-ai/mcpb`
  9. MCP Bundles (`.mcpb`) are zip archives containing a local MCP server and a `manifest.json` that describes the server and its capabilities. The format is spiritually similar to Chrome extensions (`.crx`) or VS Code extensions (`.vsix`), enabling end users to install local MCP servers with a single click.
  10. This repository provides three components: The bundle specification in [MANIFEST.md](MANIFEST.md), a CLI tool for creating bundles (see [CLI.md](CLI.md)), and the code used by Claude for macOS and Windows to load and verify MCPB bundles ([src/index.ts](src/index.ts)).
  11. - For developers of local MCP servers, we aim to make distribution and installation of said servers convenient
  12. - For developers of apps supporting local MCP servers, we aim to make it easy to add support for MCPB bundles
  13. Claude for macOS and Windows uses the code in this repository to enable single-click installation of local MCP servers, including a number of end user-friendly features - such as automatic updates, easy configuration of MCP servers and the variables and parameters they need, and a curated directory. We are committed to the open ecosystem around MCP servers and believe that its ability to be universally adopted by multiple applications and services has benefits developers aiming to connect AI tools to other apps and services. Consequently, we're open-sourcing the MCP Bundle specification, toolchain, and the schemas and key functions used by Claude for macOS and Windows to implement its own support of MCP Bundles. It is our hope that the `mcpb` format doesn't just make local MCP servers more portable for Claude, but other AI desktop applications, too.
  14. # For Bundle Developers
  15. At the core, MCPB are simple zip files containing your entire MCP server and a `manifest.json`. Consequently, turning a local MCP server into a bundle is straightforward: You just have to put all your required files in a folder, create a `manifest.json`, and then create an archive.
  16. To make this process easier, this package offers a CLI that helps you with the creation of both the `manifest.json` and the final `.mcpb` file. To install it, run:
  17. ```sh
  18. npm install -g @anthropic-ai/mcpb
  19. ```
  20. 1. In a folder containing your local MCP server, run `mcpb init`. This command will guide you through the creation of a `manifest.json`.
  21. 2. Run `mcpb pack` to create a `mcpb` file.
  22. 3. Now, any app implementing support for MCPB can run your local MCP server. As an example, open the file with Claude for macOS and Windows to show an installation dialog.
  23. You can find the full spec for the `manifest.json` and all its mandatory and optional fields in [MANIFEST.md](MANIFEST.md). Examples for bundles can be found in [examples](./examples/).
  24. ## Prompt Template for AI Tools
  25. AI tools like Claude Code are particularly good at creating MCP bundles when informed about the spec. When prompting an AI coding tool to build a bundle, briefly explain what your bundle aims to do - then add the following context to your instructions.
  26. > I want to build this as a MCP Bundle, abbreviated as "MCPB". Please follow these steps:
  27. >
  28. > 1. **Read the specifications thoroughly:**
  29. > - https://github.com/anthropics/mcpb/blob/main/README.md - MCPB architecture overview, capabilities, and integration
  30. > patterns
  31. > - https://github.com/anthropics/mcpb/blob/main/MANIFEST.md - Complete bundle manifest structure and field definitions
  32. > - https://github.com/anthropics/mcpb/tree/main/examples - Reference implementations including a "Hello World" example
  33. > 2. **Create a proper bundle structure:**
  34. > - Generate a valid manifest.json following the MANIFEST.md spec
  35. > - Implement an MCP server using @modelcontextprotocol/sdk with proper tool definitions
  36. > - Include proper error handling, security measures, and timeout management
  37. > 3. **Follow best development practices:**
  38. > - Implement proper MCP protocol communication via stdio transport
  39. > - Structure tools with clear schemas, validation, and consistent JSON responses
  40. > - Make use of the fact that this bundle will be running locally
  41. > - Add appropriate logging and debugging capabilities
  42. > - Include proper documentation and setup instructions
  43. > 4. **Test considerations:**
  44. > - Validate that all tool calls return properly structured responses
  45. > - Verify manifest loads correctly and host integration works
  46. >
  47. > Generate complete, production-ready code that can be immediately tested. Focus on defensive programming, clear error messages, and following the exact MCPB specifications to ensure compatibility with the ecosystem.
  48. ## Directory Structures
  49. ### Minimal Bundle
  50. A `manifest.json` is the only required file.
  51. ### Example: Node.js Bundle
  52. ```
  53. bundle.mcpb (ZIP file)
  54. ├── manifest.json # Required: Bundle metadata and configuration
  55. ├── server/ # Server files
  56. │ └── index.js # Main entry point
  57. ├── node_modules/ # Bundled dependencies
  58. ├── package.json # Optional: NPM package definition
  59. ├── icon.png # Optional: Bundle icon
  60. └── assets/ # Optional: Additional assets
  61. ```
  62. ### Example: Python Bundle
  63. ```
  64. bundle.mcpb (ZIP file)
  65. ├── manifest.json # Required: Bundle metadata and configuration
  66. ├── server/ # Server files
  67. │ ├── main.py # Main entry point
  68. │ └── utils.py # Additional modules
  69. ├── lib/ # Bundled Python packages
  70. ├── requirements.txt # Optional: Python dependencies list
  71. └── icon.png # Optional: Bundle icon
  72. ```
  73. ### Example: Binary Bundle
  74. ```
  75. bundle.mcpb (ZIP file)
  76. ├── manifest.json # Required: Bundle metadata and configuration
  77. ├── server/ # Server files
  78. │ ├── my-server # Unix executable
  79. │ ├── my-server.exe # Windows executable
  80. └── icon.png # Optional: Bundle icon
  81. ```
  82. ### Language Choice Recommendation
  83. **We recommend implementing MCP servers in Node.js** rather than Python to reduce installation friction. Node.js ships with Claude for macOS and Windows, which means your bundle will work out-of-the-box for users without requiring them to install additional Python runtimes (or you to package them manually).
  84. ### Bundling Dependencies
  85. **UV Runtime (v0.4+):**
  86. - Use `server.type = "uv"` in manifest
  87. - Include `pyproject.toml` with dependencies (no bundled packages needed)
  88. - Host application manages Python and dependencies automatically
  89. - Works cross-platform without user Python installation
  90. - See `examples/hello-world-uv`
  91. **Python Bundles (Traditional):**
  92. - Use `server.type = "python"` in manifest
  93. - Bundle all required packages in `server/lib/` directory
  94. - OR bundle a complete virtual environment in `server/venv/`
  95. - Use tools like `pip-tools`, `poetry`, or `pipenv` to create reproducible bundles
  96. - Set `PYTHONPATH` to include bundled packages via `mcp_config.env`
  97. - **Limitation**: Cannot portably bundle compiled dependencies (e.g., pydantic, which the MCP Python SDK requires)
  98. **Node.js Bundles:**
  99. - Run `npm install --production` to create `node_modules`
  100. - Bundle the entire `node_modules` directory with your bundle
  101. - Use `npm ci` or `yarn install --frozen-lockfile` for reproducible builds
  102. - Server entry point specified in manifest.json's `server.entry_point`
  103. **Binary Bundles:**
  104. - Static linking preferred for maximum compatibility
  105. - Include all required shared libraries if dynamic linking used
  106. - Test on clean systems without development tools
  107. # Contributing
  108. We welcome contributions! Please see our [Contributing Guide](CONTRIBUTING.md) for details.
  109. ## Development Setup
  110. ```sh
  111. # Clone the repository
  112. git clone https://github.com/anthropics/mcpb.git
  113. cd mcpb
  114. # Install dependencies
  115. yarn
  116. # Build the project
  117. yarn build
  118. # Run tests
  119. yarn test
  120. ```
  121. ## Release Process
  122. 1. Update version in `package.json`
  123. 2. Create a pull request with version bump
  124. 3. After merge, create a GitHub release
  125. 4. Package will be automatically published to npm
  126. # License
  127. This project is licensed under the Apache License 2.0 for new contributions, with existing code under MIT - see the [LICENSE](LICENSE) file for details.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.