MCP Protocol Reference

QANATIX exposes an MCP (Model Context Protocol) server via Streamable HTTP at /mcp/.

Endpoint

POST https://api.qanatix.com/mcp/

Standard MCP Streamable HTTP transport. Compatible with any MCP client (Claude Desktop, Claude Code, Cursor, custom clients).

Authentication

Include your API key in the Authorization header:

Authorization: Bearer sk_live_abc123...

In development mode, you can use X-Tenant-Id instead.

Tools

qanatix_search

Search across a vertical with hybrid matching. Returns compact markdown table by default.

Parameter	Type	Required	Default	Description
`query`	string	yes	—	Natural language search query
`vertical`	string	yes	—	Vertical to search
`filters`	object	no	—	Key-value filters on vertical_data
`limit`	integer	no	10	Results to return (1-20)

Note: MCP limit is capped at 20 (vs 100 for REST API) to keep responses within LLM context limits.

Example tool call:

{
  "name": "qanatix_search",
  "arguments": {
    "vertical": "manufacturing",
    "query": "stainless M8 bolt ISO 4017",
    "filters": {"in_stock": true},
    "limit": 5
  }
}

Returns a markdown table with name, score, and key data fields.

qanatix_lookup

Get a specific entity by UUID.

Parameter	Type	Required	Description
`entity_id`	string	yes	Full entity UUID

Returns YAML-formatted entity with all fields.

qanatix_verticals

List all available verticals with entity counts and schema status. No parameters.

Returns a markdown table of verticals.

Resources

entity://{entity_id}

Read-only access to a single entity by UUID.

entity://550e8400-e29b-41d4-a716-446655440000

Returns the same output as qanatix_lookup.

schema://{vertical}/{entity_type}

Read-only access to a vertical's JSON Schema definition.

schema://manufacturing/fastener

Returns the JSON Schema as formatted JSON.

MCP vs REST

	MCP	REST API
Transport	Streamable HTTP	Standard HTTP
Auth	Bearer token	Bearer token
Format	Compact markdown (auto)	JSON/compact/YAML (configurable)
Search limit	1-20	1-100
Reranking	Disabled by default	Enabled by default
Best for	Claude, Cursor, MCP clients	GPT, Gemini, LangChain, apps

The MCP server is optimized for AI consumption — compact output, lower limits, no reranking overhead.

MCP Protocol Reference

On this page