flutterdocs_mcp 0.1.2

Flutter/Dart API documentation for AI assistants and tools.

Flutter/Dart API Documentation for AI assistants

Wraps the offline Flutter/Dart API documentation in an MCP server that AI assistants can navigate and search.

Features

• Stores documentation in a local sqlite3 database file for fast response.
• Defines a resource URI scheme that AI assistants can predict and navigate.
• Supports both targeted documentation retrieval and full-text queries.

Getting Started

Install the latest version of flutterdocs_mcp.

dart install flutterdocs_mcp

The install command will print a message if you need to add the install directory to your shell's path.

Fetch the latest version of the documentation database file using curl or wget.

curl -L -O https://github.com/smoyerx/flutterdocs_mcp/releases/latest/download/flutterdocs.db.gz

wget --content-disposition https://github.com/smoyerx/flutterdocs_mcp/releases/latest/download/flutterdocs.db.gz

Unzip the database file and place it in a directory accessible by flutterdocs_mcp, which takes the path to the file as a command-line argument.

gunzip flutterdocs.db.gz

Verifying Versions

The version of the flutterdocs_mcp executable must match that of the database file.

To check the executable version run:

flutterdocs_mcp --version

To check the database file version run:

flutterdocs_mcp --db-version --db /path/to/flutterdocs.db

Usage

See examples and best practices for configuring flutterdocs_mcp with various hosts (e.g., GitHub Copilot in VS Code) and using it effectively.

Documentation Database

The documentation database file is generated from the Flutter stable offline documentation as follows:

• HTML content is convert to markdown and sanitized.
• Internal documentation links are converted to a custom URI scheme.
• Image links are converted to text indicating the image is omitted.
• DartPad interactive code links are converted to text indicating the interactive code is omitted.
• Stub entries are generated for inherited members to keep the URI scheme robust.
• Code examples (snippets), where available, are appended as a new section of the documentation for classes, mixins, etc.

The changelog specifies the snapshot date of the Flutter offline documentation included with a given flutterdocs_mcp release.

Motivation

There are several other MCP server projects for Flutter/Dart documentation, including:

• Context7, which is a widely-used MCP documentation service that includes various Flutter documentation sets in its inventory.
• flutter-mcp, which is built with the FastMCP Python library and fetches/caches online Flutter documentation.
• flutter_mcp_2, which is built with JavaScript and also fetches/caches online Flutter documentation.

My motivation for creating flutterdocs_mcp was to:

• Store all documentation locally for fast full-text search across all libraries.
• Eliminate first-time load delays.
• Avoid in-line conversion from HTML to markdown.
• Implement in Dart using dart_mcp for native hosting on pub.dev.

Platforms

I have only tested flutterdocs_mcp in limited configurations:

• Platforms: Ubuntu 24.04
• Hosts: GitHub Copilot in VS Code, MCP Inspector

MCP Server Operational Details

Below is a summary of how flutterdocs_mcp operates as an MCP server. You do not need to know any of this to use it, but it's here for those who are interested.

URI Scheme

• Library documentation: flutter-docs://api/{library_slug}
• Entity (class, mixin, etc.) documentation: flutter-docs://api/{library_slug}/{entity}
• Member (constructor, properties, etc.) documentation: flutter-docs://api/{library_slug}/{entity}/{member}

Library slugs are used in place of library display names to form valid URIs (e.g., dart:io uses slug dart-io). The slugs are the same as used for the native HTML documentation. AI assistants can access the mapping from library display name to slug using the listLibraries tool.

Tools

• listLibraries: lists slugs and display names for all available library documentation.
• searchDocumentation: performs a full-text search across all the library documentation.
• lookupEntity: resolves entity (class, mixin, etc.) identifier names.
• lookupMember: resolves member (constructor, property, method, etc.) identifier names.
• getDocumentation: fetches documentation for the specified URI; returns same result as the corresponding resource template.

The lookupEntity and lookupMember tools return arrays of [library_slug, entity, category] or [library_slug, entity, member, category] tuples, respectively, for constructing URIs to use with the getDocumentation tool or resource templates.

Resource Templates

There is a resource template for each of the URI shapes above: libraryIndex, entityDocumentation, and memberDocumentation. They can be used instead of the getDocumentation tool.

Not all MCP hosts support resource templates, which is why the getDocumentation tool is also provided.

TODOs

• Develop A/B tests to quantify the benefits AI assistants derive from having local access to the documentation.
• Experiment with additional tools and agent skill enhancements to increase the efficiency and effectiveness of documentation search, navigation, and retrieval.
• Add Flutter/Dart guides within a new flutter-docs://guide/... URI space.
• Test with a wider range of platforms and hosts.
• Support pagination for searchDocumentation.
• Automate the database file download via the data_assets package.

LLMs are being released or updated at an accelerating rate, so it's an open question as to how much having the most up-to-date documentation improves the performance of AI assistants. Hence why I am prioritizing A/B testing and related experimentation over additional features.

Contributing

Contributions are always appreciated and welcome! Please see these details on how you can help.

I am particularly interested in data regarding the efficacy of flutterdocs_mcp or any other MCP server that wraps Flutter/Dart documentation.

Acknowledgements

• Flutter/Dart team for making their excellent documentation available offline, and for providing the dart_mcp package which makes building MCP servers a breeze.
• Developers of the html-to-markdown package which has proven robust and blazingly fast.
• Sonnet 4.5/4.6 for writing most of the code. Yes, I reviewed it (OK, the tests not so much).