Directree README | AllMCPs

directree MCP Server

Query directree — the honest software directory — from any AI assistant that supports the Model Context Protocol.

Search 5,000+ software tools, get structured facts with clear provenance labels (observed / AI-inferred / founder-edited), and compare products side by side.

Tools

Tool	Description
`search_tools`	Search by name, tagline, or description. Filter by category or free plan.
`get_tool`	Get full structured facts for one tool by slug.
`compare_tools`	Compare 2-4 tools side by side.
`list_categories`	List all software categories with tool counts.

Usage

Option A: Remote HTTP endpoint (no install)

Any MCP-compatible client can connect to the hosted endpoint directly:

Code

https://www.directree.io/api/mcp

Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

config.json

{
  "mcpServers": {
    "directree": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://www.directree.io/api/mcp"]
    }
  }
}

Restart Claude Desktop after saving. The directree tools will appear in the tools panel.

Option B: Run locally (stdio)

bash

# Clone and install
git clone https://github.com/coburn2716/directree-mcp.git
cd directree-mcp
npm install

# Run
node mcp-server.mjs

No API key required. Reads directly from the directree public catalog.

Claude Desktop config for local:

config.json

{
  "mcpServers": {
    "directree": {
      "command": "node",
      "args": ["/path/to/directree-mcp/mcp-server.mjs"]
    }
  }
}

REST API

Prefer plain REST? Use the agent-ready endpoints directly:

bash

# Search tools
curl "https://www.directree.io/api/agent/tools?q=project+management&has_free_plan=true"

# Get one tool
curl "https://www.directree.io/api/agent/tools/notion"

# List categories
curl "https://www.directree.io/api/agent/categories"

Full docs: directree.io/mcp

Why directree?

Every fact in the directory is labelled with its source:

Observed - crawled directly from the tool's own website
AI-inferred - generated by a model, clearly marked as such
Founder-edited - verified and corrected by the tool's owner

No fake scores. No pay-to-rank. Clean, structured data agents can actually trust.

License

MIT