Skip to content
Wexa Fabric

Connect Fabric as an MCP server

Fabric is a remote MCP server. Any MCP-capable agent — Claude.ai, Claude Desktop, Cursor, ChatGPT, an IDE assistant, a CI job, or your own code — connects to Fabric and gains governed access to your enterprise context graph. The agent calls Fabric’s tools; Fabric grounds the answer in your real data and records every call.

This is the most common way to integrate with Fabric, and the focus of this section.

The “Connect your agent” tabs (MCP / API / SDK) with the MCP client picker.

How it works

Section titled “How it works”
  1. You generate a project-scoped API key (fab_sk_…) or authorize via OAuth 2.1.
  2. You point your MCP client at the Fabric MCP Server URL and pass your key and workspace ID as headers.
  3. Your agent discovers Fabric’s tools (query-context, create-ontology, save-context, docs) and calls them like any other MCP tools.
  4. Every call runs Fabric’s governed pipeline — scope-pinned, validated, policy-checked, executed, redacted, and audited — and returns a lifecycle_id.

The context graph itself is never exposed. Your queries are pinned to your project, read-only by default, and bounded — so an agent can never read another tenant’s data.

Transports

Section titled “Transports”

Fabric supports both MCP transports, so it works with hosted clients and local clients alike:

Transport Endpoint Use it for
Streamable HTTP POST /mcp Modern MCP clients (the default)
SSE (compatibility) GET /sse + POST /messages?session=… Clients that connect over Server-Sent Events

Your console shows the exact URL for your deployment (for example https://mcp.wexa.ai/v1/sse). Use that value.

Pick your client

Section titled “Pick your client”
Client Guide
Claude.ai (custom connector) and Claude Desktop (config file) Claude
Cursor Cursor
ChatGPT (Developer mode) ChatGPT
Any other MCP client Other MCP clients

Before you start

Section titled “Before you start”