New · Model Context Protocol

Talk to your product data
inside your IDE.

Inslytic exposes a remote MCP server. Point Claude Desktop, Claude Code, Cursor, or any MCP-compatible client at it — and ask real questions about real users in the same chat where you write code.

Remote HTTP transportBearer token authRead-only, project-scoped

What your agent can do

A small surface, on purpose. Two tools is enough — agents are good at SQL once you give them the schema.

Read your event taxonomy

Agents call get_event_taxonomy first to learn your product description, event definitions, and the names of every event you actually track.

Run scoped SQL queries

query_events accepts read-only ClickHouse SQL with placeholders for project ID and time range. The server injects values — no cross-project access, ever.

EU-resident, never trains a model

Queries hit Inslytic's EU infrastructure under your existing project API key. Same data residency story as the rest of the product.

Setup

Two steps. Pick the snippet that matches your client.

Grab your API key

Open the Inslytic dashboard → Settings. Copy the project API key — it's the same key the Ingestion API uses, scoped to a single project.

Add the server to your MCP client

Pick the snippet that matches your client.

Claude Desktop · claude_desktop_config.json
{
  "mcpServers": {
    "inslytic": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.inslytic.com/v1/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "AUTH_HEADER": "Bearer ins_your_api_key_here"
      }
    }
  }
}
Claude Code · one command
claude mcp add --transport http inslytic https://api.inslytic.com/v1/mcp \
  --header "Authorization: Bearer ins_your_api_key_here"
Cursor & native HTTP MCP · mcp.json
{
  "mcpServers": {
    "inslytic": {
      "url": "https://api.inslytic.com/v1/mcp",
      "headers": {
        "Authorization": "Bearer ins_your_api_key_here"
      }
    }
  }
}

Replace ins_your_api_key_here with your real key, restart your client, and Inslytic will show up as a connected tool.

The two tools

Both read-only. Both project-scoped. That's the whole API.

get_event_taxonomyread

Returns the project's event definitions, product description, and every event name actually present in your stream. Agents call this first so they query against real fields, not hallucinated ones.

{
  "productDescription": "B2B SaaS dashboard for marketing teams.",
  "events": [
    { "name": "signup_completed",     "category": "activation"   },
    { "name": "subscription_started", "category": "monetization" },
    { "name": "report_exported",      "category": "engagement"   }
  ],
  "trackedEventNames": ["signup_completed", "page_viewed", ...]
}
query_eventsread

Executes read-only ClickHouse SQL against your events table. The agent writes the query with placeholders; the server injects {projectId:UUID}, {from:DateTime64(3)}, and {to:DateTime64(3)} so queries are always scoped to your project.

SELECT event_name, count() AS events
FROM events
WHERE project_id = {projectId:UUID}
  AND timestamp BETWEEN {from:DateTime64(3)} AND {to:DateTime64(3)}
GROUP BY event_name
ORDER BY events DESC
LIMIT 20;

Things to know

Transport

JSON-RPC 2.0 over HTTP at POST /v1/mcp. Use mcp-remote for clients that only speak stdio.

Auth

Bearer token. Same project API key the Ingestion API uses. Revoke or rotate from the dashboard.

Scope

Read-only. Queries run as a ClickHouse read-only user, scoped to your project ID. No mutations, no cross-tenant data, no surprises.

Plan availability

Free during the public preview. Usage will count toward your AI query allowance once we launch billing.

Ship the question, not the export.

Connect MCP in two minutes and start asking your AI agent questions about real users, real events.