Voiceflow Docs MCP Server

An MCP (Model Context Protocol) server that gives AI assistants searchable access to the complete Voiceflow documentation. Built for developers building with Voiceflow who want their AI tools to have full context on the platform.

Features

• Search across all Voiceflow docs with BM25 ranking
• Fetch complete page content by path
• Browse the documentation structure
• Always fresh — dynamically fetches and caches docs (24h refresh)
• Works offline — falls back to cached data when network is unavailable
• Zero config — just add to your MCP client and go

Installation

Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "voiceflow-docs": {
      "command": "npx",
      "args": ["-y", "voiceflow-docs-mcp"]
    }
  }
}

Claude Code

claude mcp add voiceflow-docs -- npx -y voiceflow-docs-mcp

Cursor

Add to your Cursor MCP config (.cursor/mcp.json in your project root):

{
  "mcpServers": {
    "voiceflow-docs": {
      "command": "npx",
      "args": ["-y", "voiceflow-docs-mcp"]
    }
  }
}

Other MCP Clients

Run as a stdio MCP server:

npx voiceflow-docs-mcp

Tools

search_voiceflow_docs

Search across all Voiceflow documentation by keyword or question.

| Parameter | Type | Description | |-----------|------|-------------| | query | string | Search query — keywords or natural language | | section | string? | Filter: documentation, api-reference, courses, developer-tools, changelog, trace-types | | max_results | number? | Max results (default 10, max 25) |

fetch_voiceflow_page

Retrieve the complete content of a documentation page by path.

| Parameter | Type | Description | |-----------|------|-------------| | path | string | Page path, e.g. /documentation/build/tools/api |

list_voiceflow_sections

Browse the documentation structure and navigation tree.

| Parameter | Type | Description | |-----------|------|-------------| | section | string? | Filter to a specific section (optional) |

How It Works

On first run, the server fetches Voiceflow's llms-full.txt (~150KB of pre-formatted documentation), parses it into pages, chunks content by heading boundaries, and builds a BM25 search index. The processed data is cached locally at ~/.voiceflow-docs-mcp/cache.json.

Subsequent startups load from cache in under 100ms. The cache auto-refreshes every 24 hours. Pass --refresh to force an immediate re-fetch.

Testing

Unit Tests

npm test

Runs parser, chunker, and BM25 search tests against fixture data.

MCP Inspector (Interactive Testing)

The MCP Inspector provides a visual UI to test your MCP tools interactively.

# Test the built version
npm run build
npm run inspect

# Test from source (no build needed)
npm run inspect:dev

This opens a browser UI where you can:

• See all 3 tools listed with their schemas
• Call search_voiceflow_docs with any query and see ranked results
• Call fetch_voiceflow_page with a path and see full page content
• Call list_voiceflow_sections to browse the doc tree

Test with Claude Code (Local)

Add the server locally pointing to the built file:

# Build first
npm run build

# Add as local MCP server
claude mcp add voiceflow-docs-local -- node /path/to/voiceflow-docs-mcp/dist/index.js

# Or from source (no build needed)
claude mcp add voiceflow-docs-local -- npx tsx /path/to/voiceflow-docs-mcp/src/index.ts

Then in a Claude Code conversation, ask something like:

• "Search the Voiceflow docs for knowledge base"
• "How do I set up a web chat widget in Voiceflow?"
• "List all Voiceflow documentation sections"

Test with Claude Desktop (Local)

Add to your Claude Desktop config pointing to the local build:

{
  "mcpServers": {
    "voiceflow-docs": {
      "command": "node",
      "args": ["/absolute/path/to/voiceflow-docs-mcp/dist/index.js"]
    }
  }
}

Restart Claude Desktop and the tools will appear.

Development

# Install dependencies
npm install

# Run from source
npm run dev

# Build for distribution
npm run build

# Test with MCP Inspector
npm run inspect

# Run tests
npm test

# Type check
npm run typecheck

Caching

• Location: ~/.voiceflow-docs-mcp/cache.json
• TTL: 24 hours — auto-refreshes when stale
• Offline: Falls back to stale cache if the network is unavailable
• Force refresh: Pass --refresh flag (e.g. npx voiceflow-docs-mcp --refresh)
• Clear cache: Delete ~/.voiceflow-docs-mcp/cache.json

Requirements

• Node.js 18+

License

MIT