Provides AI-powered synthesis of documentation responses using Google Gemini's API as one of the supported AI provider options for the SMRT framework documentation assistant.
Provides AI-powered synthesis of documentation responses using OpenAI's API as one of the supported AI provider options for the SMRT framework documentation assistant.
@happyvertical/smrt-docs-mcp
Documentation MCP server for the SMRT framework that acts as an orchestrator, routing developer queries to appropriate package experts using CLAUDE.md files.
Overview
The SMRT Documentation MCP Server implements a RAG (Retrieval-Augmented Generation) pattern where each SMRT package's CLAUDE.md file serves as domain expertise. When you ask a question, the server:
Routes your query to relevant packages based on keyword matching
Loads the CLAUDE.md documentation for those packages
Uses AI to synthesize a response based on the expert documentation
Returns an answer with package references
Installation
Usage
As an MCP Server
Add to your .mcp.json (see Configuration section below):
Environment Variables
The ask tool requires an AI provider to be configured. Set one of:
HAVE_AI_API_KEY- Fallback API key for any providerHAVE_AI_TYPE- Provider type ('openai', 'anthropic', 'gemini')OPENAI_API_KEY- OpenAI API keyANTHROPIC_API_KEY- Anthropic API keyGEMINI_API_KEY- Google Gemini API key
Other tools (list-packages, get-docs) work without AI configuration.
Available Tools
ask
Ask a question about the SMRT framework. Automatically routes to relevant packages and synthesizes a response.
Input:
Example:
list-packages
List all available SMRT framework packages with descriptions and keywords.
Input:
Example Output:
get-docs
Get the full CLAUDE.md documentation for a specific package.
Input:
Example:
How It Works
Package Registry
The registry scans packages/*/CLAUDE.md files at startup and builds a catalog of available packages with:
Package name
Description (extracted from CLAUDE.md)
Keywords for routing
Full documentation content
Query Routing
When you ask a question, the router:
Extracts keywords from your query
Matches against package keyword lists
Scores packages by relevance
Returns top matches
AI Synthesis
The ask tool:
Loads CLAUDE.md for relevant packages (top 3 matches)
Builds context from documentation
Uses AI (via
@happyvertical/ai) to generate responseIncludes package references in response
Package Keywords
Each package has associated keywords for routing. Examples:
core: smrt, framework, orm, database, collection, object, decorator, schema, api, rest, cli, mcp
agents: agent, autonomous, actor, workflow, orchestration, task
content: content, article, markdown, cms, document, blog, publishing
events: event, calendar, meeting, schedule, attendance, rsvp
gnode: federation, network, knowledge base, distributed, peer
See src/registry.ts for the complete mapping.
Development
Configuration
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
Claude Code
Add to project root .mcp.json:
Adding New Packages
When you add a new package to the SMRT framework:
Create a
CLAUDE.mdfile documenting the packageUpdate
src/registry.tsto add keywords for the new packageRebuild:
pnpm run build
The package will be automatically discovered and included in the registry.
Architecture
Troubleshooting
"AI client initialization failed"
The ask tool requires an AI provider. Set one of the environment variables listed above.
"No relevant packages found"
Try using list-packages to see available packages, or specify packages explicitly with the packages parameter.
"Package not found"
Use list-packages to see the exact package names. Package names should not include the @happyvertical/smrt- prefix (use "core" not "smrt-core").
License
MIT
Related
SDK MCP Server - Similar implementation for the SDK