Corvi MCP Server

How to use

You can connect your MCP client to either a hosted MCP endpoint or to a local proxy that relays to a hosted MCP. The available configurations show both a direct HTTP connection and a local proxy command that spawns an MCP proxy using a standard package runner.

How to install

Prerequisites depend on which integration you choose. You will typically need Node.js version 18 or newer for npx-based workflows, and Python 3.11+ if you plan to use the Python CLI. Install steps below assume you want to use the hosted MCP endpoint and a local proxy as shown in practical examples.

# Install Node.js 18+ if you haven’t already
# Use your system’s package manager or install from nodejs.org

# Verify Node.js installation
node -v
npm -v

# Optional: install Python 3.11+ if you will run the Python CLI
python3 --version

Additional setup steps

If you want Claude Desktop to route through a local MCP proxy, you can configure Claude to use a local definition that points to a hosted MCP endpoint. This requires adding a local proxy definition to Claude’s config and then restarting Claude Desktop to recognize the new server.

Additional sections

This MCP server supports multiple connection methods. Use the HTTP method to connect directly to the hosted MCP endpoint, or use a local stdio-based proxy to launch an MCP client that talks to the hosted endpoint. If you plan to operate via code, you can interact with the MCP server through standard HTTP or via local processes that spawn MCP-interacting commands.

Configuration examples

Use the following configuration blocks to connect to the hosted MCP endpoint. The HTTP option connects directly to the hosted MCP URL, while the STDIO option runs a local proxy command that forwards to the hosted MCP.

{
  "mcpServers": {
    "corvi_proxy": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote@latest",
        "https://corvi.careers/mcp"
      ]
    }
  }
}

OpenAI integration example

If you are using OpenAI tools to interact with the MCP server, you will point your client to a streamable HTTP MCP endpoint and define your agent to use the available MCP tools to perform tasks such as location autocomplete and job searches.

import asyncio
from agents import Agent, Runner
from agents.mcp.server import MCPServerStreamableHttp
import json

MCP_URL = "https://jobswithgpt.com/mcp"  # your FastMCP streamable HTTP endpoint

async def main():
    async with MCPServerStreamableHttp(params={"url": MCP_URL}, name="corvi.careers") as server:
        agent = Agent(
            name="jobs-mcp-local",
            mcp_servers=[server],
            instructions=(
                "Use the MCP server tools. First call location_autocomplete to get a geonameid "
                "for 'Seattle', then call search_jobs with keywords=['python'] and that geonameid."
            ),
        )
        res = await Runner.run(agent, "Find machine learning jobs in san francisco.")
        print(res.final_output)

if __name__ == "__main__":
    asyncio.run(main())

Local Job Search CLI

A small Python CLI, managed by uv, lets you chat with the hosted MCP server directly from your terminal. It remembers your session so you can run multiple queries in one go.

uv sync  # creates a virtual environment in .venv and installs httpx
```
```bash
uv run jobs-mcp

Usage during a session

While the session is active, you can type natural language queries such as “python jobs in Seattle” or use special commands like schema to print the MCP tool descriptions, or quit to exit.

python jobs in Seattle
schema
quit

Notes on tooling and endpoints

Key tools exposed by the MCP server include autocompletion for locations and a job search capability. These tools enable you to discover geonames and fetch job listings through the MCP protocol.

location_autocomplete

Provides location suggestions and geoname identifiers to support contextual searches.

search_jobs

Performs searchable queries over job postings using provided keywords and location context.