Connect any other MCP client

Any MCP-compatible client can connect to Fabric — coding assistants, chat agents, IDE tools, CI systems, or your own MCP client. The pattern is always the same: point the client at Fabric’s MCP URL and pass two headers.

First, get your MCP credentials: your MCP Server URL, workspace ID (ws_<your_project>), and an API key (fab_sk_…).

The connection pattern

1. Point your client at the Server URL from your console.
2. Pass the Authorization and Workspace headers with every request.
3. If your client uses a config file instead of fields, use the JSON form below.

Header	Value
Authorization	Bearer <YOUR_API_KEY>
X-Wexa-Workspace	ws_<your_project>

Config-file clients

mcp.json

{
  "mcpServers": {
    "wexa-context": {
      "url": "https://mcp.wexa.ai/v1/sse",
      "headers": {
        "Authorization": "Bearer <YOUR_API_KEY>",
        "X-Wexa-Workspace": "ws_<your_project>"
      }
    }
  }
}

Choosing a transport

Transport	Endpoint	When to use
Streamable HTTP	POST /mcp	Modern MCP clients (default)
SSE (compat)	GET /sse + POST /messages?session=…	Clients that speak Server-Sent Events

Fabric negotiates the MCP protocol version with your client automatically. Both transports run the same governed pipeline and expose the same tools.

OpenAI custom tools (SSE)

For OpenAI custom tools, use the /sse URL with OAuth, or pass the Authorization and workspace headers with an API key. The tools your agent sees are the Fabric MCP tools documented in The Fabric MCP tools.

Verify

Issue a tools/list from your client — you should see query-context, docs, and (depending on mode) create-ontology and save-context. Then call query-context with a natural-language query.

Next steps

• The Fabric MCP tools
• Authentication & scopes