MCP API

The CF Benchmarks Model Context Protocol (MCP) server lets AI assistants such as Claude and Codex query CF Benchmarks data directly, using the same API key as the REST and WebSocket APIs. Once connected, the assistant can look up historical index values, retrieve index methodology, and fetch asset supply and metadata through the tools described in this section.

Internal testing

This endpoint is for internal testing only and is not yet publicly reachable. A public beta endpoint will be made available later. Please contact CF Benchmarks for access.

Endpoint

https://<internal-test-environment>/sse

The server communicates over HTTP using Server-Sent Events (SSE), the standard MCP remote transport.

Authentication

The MCP server uses the same API key as the REST and WebSocket APIs. This key can be obtained by contacting CF Benchmarks for a license.

Authentication uses HTTP Basic auth: provide an Authorization header of the form Authorization: Basic <credentials>, where <credentials> is the Base64 encoding of your username and key joined by a single colon : (i.e. base64("username:key")).

Individual tools are further gated by the permissions attached to your API key — for example, a tool will return an authorization error if your key is not licensed for the requested index or dataset.

Connecting a client

Most MCP clients can connect to a remote SSE server that requires an Authorization header. The examples below assume <credentials> is the Base64 string described above.

Claude Code

claude mcp add --transport sse cf-benchmarks https://<internal-test-environment>/sse \
  --header "Authorization: Basic <credentials>"

Claude Desktop

Add the following to your claude_desktop_config.json (Settings → Developer → Edit Config). It uses the mcp-remote bridge to forward the auth header:

{
  "mcpServers": {
    "cf-benchmarks": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://<internal-test-environment>/sse",
        "--header",
        "Authorization: Basic <credentials>"
      ]
    }
  }
}

Codex

Add the following to ~/.codex/config.toml:

[mcp_servers.cf-benchmarks]
command = "npx"
args = [
  "-y",
  "mcp-remote",
  "https://<internal-test-environment>/sse",
  "--header",
  "Authorization: Basic <credentials>",
]

After connecting, restart the client and confirm the cf-benchmarks tools are listed.

Available tools

Tool	Description
get_recent_results	Recent index values as a time series from an in-memory cache (no Trino) — roughly the last year for RR, last hour for RTI. Prefer over get_historical_results for recent ranges.
get_historical_results	Historical publishable index values for one or more (index, datetime) lookups.
get_index_info	Index methodology and information from cfbenchmarks.com.
get_supplies	CF Benchmarks asset supply data.
get_asset_metadata	Asset identity, DACS classification, related indices, and basket membership.
get_factors_data	CF Benchmarks factor scores, betas, and returns.

For recent data, prefer get_recent_results — it serves from memory and avoids querying Trino; use get_historical_results only for older or exact-point history.