Skip to main content
mcp-hyperspeed is a stdio JSON-RPC MCP server that proxies tool calls from an AI coding assistant to the Hyperspeed API. It lets Cursor, Claude Desktop, or any other MCP-compatible client invoke Hyperspeed’s agent tools — such as reading files, searching chat, and proposing patches — directly from the AI’s tool-calling loop.

Requirements

You need three environment variables before starting the server:
The server exits immediately if any of these variables are missing or empty.

Running the server

1

Build or run from source

From the repository root, run the server directly:
Or build a binary first:
Then run it:
2

Configure your MCP client

Add mcp-hyperspeed as a server in your MCP client configuration. See the example configs below.

MCP protocol

The server speaks stdio JSON-RPC using the Model Context Protocol (protocol version 2024-11-05). It supports: Any other method returns a method not found error. Notification messages (prefixed notifications/) are silently ignored.

Authentication

The server authenticates every API request with Authorization: Bearer <HYPERSPEED_TOKEN>. Use a service account token (sa_…) created from Workspace Settings → AI Staff. The service account must have permission to access the tools it will invoke. For most use cases, the default system roles assigned at creation are sufficient.
A 401 or 403 response from the API means the token is invalid or the service account does not have access to the requested org. Check HYPERSPEED_TOKEN and HYPERSPEED_ORG_ID.

Optional: per-call metadata

You can pass control metadata to the server by including a _hyperspeed key in any tool’s arguments object. The server strips this key before forwarding arguments to the API.
Supported fields:

Example client configuration

Use a dedicated service account for each MCP client (e.g. one for Cursor, one for Claude). This gives you per-client audit trails and lets you rotate tokens independently.