Model Context Protocol

Cortexa in every MCP client

Add Cortexa as a tool inside Claude Desktop, Cursor, Codex, Claude Code, Windsurf, Zed, and any other MCP-compatible agent. One command, one API key, one cited medical-research answer.

Get API key Local install (stdio)Remote (HTTP)

Single tool, deep answers

Your host LLM sees one tool: cortexa_research. Cortexa decomposes the question, queries live databases, cross-references sources, and returns a cited markdown answer.

Two transports, one schema

Stdio via npx for desktop IDEs, streamable-HTTP for remote and browser-based clients. Both expose the exact same tool definition.

Your Cortexa key

Auth is your existing Cortexa API key. Usage is metered under the same plan as /v1 traffic. Revoke a key and every MCP client using it disconnects immediately.

The tool

cortexa_research

Ask any research question — biomedical, clinical, genomics, drug discovery, chemistry, structures, AI, physics, engineering, materials, energy, or policy. Cortexa decomposes it, routes across 2.1K+ verified research tools, and returns a cited markdown brief.

Input schema

{
  "question": string  // 4 – 4000 characters
}

Local install (stdio)

Zero-install via npx -y @yoreai/cortexa-mcp. Your MCP client spawns a subprocess that streams JSON-RPC over stdin/stdout. Works offline for everything except the actual Cortexa call.

Claude Desktop

macOS / Windows / Linux

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) · %APPDATA%\Claude\claude_desktop_config.json (Windows)

json

{
  "mcpServers": {
    "cortexa": {
      "command": "npx",
      "args": ["-y", "@yoreai/cortexa-mcp"],
      "env": {
        "CORTEXA_API_KEY": "sk-cortexa-live-..."
      }
    }
  }
}

Quit and relaunch Claude. The tool shows up as 'cortexa_research' in the hammer icon.

Cursor

The editor you're probably reading this in

~/.cursor/mcp.json (user) · .cursor/mcp.json (workspace)

json

{
  "mcpServers": {
    "cortexa": {
      "command": "npx",
      "args": ["-y", "@yoreai/cortexa-mcp"],
      "env": { "CORTEXA_API_KEY": "sk-cortexa-live-..." }
    }
  }
}

Reload Cursor. In chat, tools appear under the 'Available Tools' list. Ask a medical question and watch Cortexa handle it.

Claude Code

Anthropic's terminal coding agent

~/.claude/settings.json (or per-project .claude.json)

json

{
  "mcpServers": {
    "cortexa": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@yoreai/cortexa-mcp"],
      "env": { "CORTEXA_API_KEY": "sk-cortexa-live-..." }
    }
  }
}

Run `claude mcp list` to confirm the server registered, then `claude mcp tools cortexa` to see cortexa_research.

OpenAI Codex CLI

Open-source terminal agent

~/.codex/config.toml

toml

[mcp_servers.cortexa]
command = "npx"
args = ["-y", "@yoreai/cortexa-mcp"]
env = { CORTEXA_API_KEY = "sk-cortexa-live-..." }

Codex picks up the tool on the next run. Try 'Ask Cortexa: what's the 2026 first-line therapy for HER2+ mBC?'

Windsurf

Codeium's AI editor

~/.codeium/windsurf/mcp_config.json

json

{
  "mcpServers": {
    "cortexa": {
      "command": "npx",
      "args": ["-y", "@yoreai/cortexa-mcp"],
      "env": { "CORTEXA_API_KEY": "sk-cortexa-live-..." }
    }
  }
}

Zed

High-performance editor

Zed settings: search 'context servers' and add Cortexa, or paste into ~/.config/zed/settings.json

json

{
  "context_servers": {
    "cortexa": {
      "command": {
        "path": "npx",
        "args": ["-y", "@yoreai/cortexa-mcp"],
        "env": { "CORTEXA_API_KEY": "sk-cortexa-live-..." }
      }
    }
  }
}

Remote (streamable-HTTP)

No local install. Point your MCP client at https://api.cortexa.sh/mcp and authenticate with a Cortexa bearer token. This is the only option for browser-based hosts (Claude.ai connectors) and container-only agents that can't run subprocesses.

Remote / HTTP clients

Claude.ai web connectors, browser-based MCP hosts, internal agents — anything that speaks streamable-HTTP MCP

Endpoint URL

json

{
  "mcpServers": {
    "cortexa": {
      "type": "http",
      "url": "https://api.cortexa.sh/mcp",
      "headers": {
        "Authorization": "Bearer sk-cortexa-live-..."
      }
    }
  }
}

Some clients ask for the endpoint and token separately — paste https://api.cortexa.sh/mcp as the URL and your key as the bearer token. No local install required.

Smoke-test the endpoint

Want to check your key is good before wiring up a client? Export your API key and run these two curls.

shell

# 1. verify auth + tool listing
curl -s https://api.cortexa.sh/mcp \
  -X POST \
  -H "Authorization: Bearer $CORTEXA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

# 2. run the agent
curl -s https://api.cortexa.sh/mcp \
  -X POST \
  -H "Authorization: Bearer $CORTEXA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0", "id": 2, "method": "tools/call",
    "params": {
      "name": "cortexa_research",
      "arguments": {
        "question": "Latest 2026 guidance on ICI-induced myocarditis management?"
      }
    }
  }'

FAQ

The tool

cortexa_research

Input schema

{
  "question": string  // 4 – 4000 characters
}

Local install (stdio)

Zero-install via npx -y @yoreai/cortexa-mcp. Your MCP client spawns a subprocess that streams JSON-RPC over stdin/stdout. Works offline for everything except the actual Cortexa call.

Claude Desktop

macOS / Windows / Linux

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) · %APPDATA%\Claude\claude_desktop_config.json (Windows)

json

{
  "mcpServers": {
    "cortexa": {
      "command": "npx",
      "args": ["-y", "@yoreai/cortexa-mcp"],
      "env": {
        "CORTEXA_API_KEY": "sk-cortexa-live-..."
      }
    }
  }
}

Quit and relaunch Claude. The tool shows up as 'cortexa_research' in the hammer icon.

Cursor

The editor you're probably reading this in

~/.cursor/mcp.json (user) · .cursor/mcp.json (workspace)

json

{
  "mcpServers": {
    "cortexa": {
      "command": "npx",
      "args": ["-y", "@yoreai/cortexa-mcp"],
      "env": { "CORTEXA_API_KEY": "sk-cortexa-live-..." }
    }
  }
}

Reload Cursor. In chat, tools appear under the 'Available Tools' list. Ask a medical question and watch Cortexa handle it.

Claude Code

Anthropic's terminal coding agent

~/.claude/settings.json (or per-project .claude.json)

json

{
  "mcpServers": {
    "cortexa": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@yoreai/cortexa-mcp"],
      "env": { "CORTEXA_API_KEY": "sk-cortexa-live-..." }
    }
  }
}

Run `claude mcp list` to confirm the server registered, then `claude mcp tools cortexa` to see cortexa_research.

OpenAI Codex CLI

Open-source terminal agent

~/.codex/config.toml

toml

[mcp_servers.cortexa]
command = "npx"
args = ["-y", "@yoreai/cortexa-mcp"]
env = { CORTEXA_API_KEY = "sk-cortexa-live-..." }

Codex picks up the tool on the next run. Try 'Ask Cortexa: what's the 2026 first-line therapy for HER2+ mBC?'

Windsurf

Codeium's AI editor

~/.codeium/windsurf/mcp_config.json

json

{
  "mcpServers": {
    "cortexa": {
      "command": "npx",
      "args": ["-y", "@yoreai/cortexa-mcp"],
      "env": { "CORTEXA_API_KEY": "sk-cortexa-live-..." }
    }
  }
}

Zed

High-performance editor

Zed settings: search 'context servers' and add Cortexa, or paste into ~/.config/zed/settings.json

json

{
  "context_servers": {
    "cortexa": {
      "command": {
        "path": "npx",
        "args": ["-y", "@yoreai/cortexa-mcp"],
        "env": { "CORTEXA_API_KEY": "sk-cortexa-live-..." }
      }
    }
  }
}

Remote (streamable-HTTP)

Remote / HTTP clients

Claude.ai web connectors, browser-based MCP hosts, internal agents — anything that speaks streamable-HTTP MCP

Endpoint URL

json

{
  "mcpServers": {
    "cortexa": {
      "type": "http",
      "url": "https://api.cortexa.sh/mcp",
      "headers": {
        "Authorization": "Bearer sk-cortexa-live-..."
      }
    }
  }
}

Some clients ask for the endpoint and token separately — paste https://api.cortexa.sh/mcp as the URL and your key as the bearer token. No local install required.

Smoke-test the endpoint

Want to check your key is good before wiring up a client? Export your API key and run these two curls.

shell

# 1. verify auth + tool listing
curl -s https://api.cortexa.sh/mcp \
  -X POST \
  -H "Authorization: Bearer $CORTEXA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

# 2. run the agent
curl -s https://api.cortexa.sh/mcp \
  -X POST \
  -H "Authorization: Bearer $CORTEXA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0", "id": 2, "method": "tools/call",
    "params": {
      "name": "cortexa_research",
      "arguments": {
        "question": "Latest 2026 guidance on ICI-induced myocarditis management?"
      }
    }
  }'

FAQ