Using the Madjik API for MCP

Built-in Model Context Protocol tool endpoints.

The API exposes MCP-format tools directly — you do not need to write a wrapper server unless you want one. See the API reference for the endpoints.

Available tools

Tool	What it does
`madjik_get_metrics`	Latest values for given ids or categories (all accessible metrics if empty)
`madjik_get_timeseries`	History for one metric (`metric_id`, `hours`, `interval`)
`madjik_market_summary`	Summary across key metrics for a focus area (`overview`, `risk`, `sentiment`, `derivatives`)
`madjik_list_metrics`	Catalog listing with tiers and accessibility

Calling a tool

# discover tools (public)
curl "https://api.madjik.io/mcp/tools"

# call one (authenticated)
curl -X POST -H "X-API-Key: YOUR_API_KEY" -H "Content-Type: application/json" \
  -d '{"name": "madjik_market_summary", "arguments": {"focus": "risk"}}' \
  "https://api.madjik.io/mcp/call"

Responses follow the MCP content format:

{ "content": [ { "type": "text", "text": "{ ...JSON payload... }" } ], "isError": false }

Bridging to an MCP client

To use these tools from an MCP-speaking assistant, run a thin stdio/HTTP bridge that lists /mcp/tools and forwards calls to /mcp/call with your API key. Keep the key in the bridge's environment — never in the model context.

Best practices

Cache tool results briefly — repeated identical calls burn your quota
Surface descriptions — pass each tool's description through to the model
Return errors as data — the endpoint sets isError: true rather than failing the transport