Setup

The Precursor Intelligence MCP server is remote: a single hosted HTTP endpoint. There is nothing to install locally. You only need:

1. An API key from your Precursor Intelligence dashboard.
2. An MCP-compatible client.

1. Create an API key

Sign in to precursorintelligence.com and go to Settings → API Keys.

1. Click Create key.
2. Name the key (e.g. claude-desktop-laptop).
3. (Optional) Add an IP allowlist.
4. Click Generate. The full token is shown once; copy it now.

2. Connect a client

Choose your client below. The configuration shape is the same everywhere: a remote HTTP MCP server with a bearer-token Authorization header, with only the file path and JSON dialect changing.

Claude Code

From a terminal:

claude mcp add --transport http precursor-intelligence \
  https://mcp.precursorintelligence.com \
  --header "Authorization: Bearer <YOUR_API_KEY>"

Verify the connection:

claude mcp list

Inside a Claude Code session you can confirm the server is live:

/mcp

Claude Desktop

Edit your Claude Desktop config:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "precursor-intelligence": {
      "type": "http",
      "url": "https://mcp.precursorintelligence.com",
      "headers": {
        "Authorization": "Bearer <YOUR_API_KEY>"
      }
    }
  }
}

Restart Claude Desktop. The 41 tools will appear under the tool menu prefixed with precursor-intelligence.

Cursor

Edit ~/.cursor/mcp.json (global) or .cursor/mcp.json (project-level). Cursor supports environment variable interpolation, so you can keep your key out of the config file:

{
  "mcpServers": {
    "precursor-intelligence": {
      "type": "http",
      "url": "https://mcp.precursorintelligence.com",
      "headers": {
        "Authorization": "Bearer ${env:PRECURSOR_API_KEY}"
      }
    }
  }
}

Restart Cursor or run MCP: Reload Servers from the command palette.

Raw JSON-RPC / custom clients

Any HTTP client can speak to the server directly. Send a POST to https://mcp.precursorintelligence.com with Content-Type: application/json and your bearer token. Use tools/list for discovery and tools/call for invocation:

curl -s https://mcp.precursorintelligence.com \
  -H "Authorization: Bearer $PRECURSOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

curl -s https://mcp.precursorintelligence.com \
  -H "Authorization: Bearer $PRECURSOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"get_cve","arguments":{"cve_id":"CVE-2024-3094"}}}'

See Response Format for the full envelope shape.

Other clients and the mcp-remote fallback

If your client does not yet support Streamable HTTP MCP servers natively, bridge through mcp-remote, a stdio shim that proxies to any remote MCP endpoint:

{
  "mcpServers": {
    "precursor-intelligence": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.precursorintelligence.com",
        "--header",
        "Authorization: Bearer <YOUR_API_KEY>"
      ]
    }
  }
}

3. Test the connection

The quickest health-check is the account_status tool. It is free and confirms auth and credit balance in one call.

From inside your client, ask:

Call the account_status tool on the Precursor Intelligence MCP server.

Or hit the endpoint directly to verify the wire-level setup:

curl -s https://mcp.precursorintelligence.com \
  -H "Authorization: Bearer $PRECURSOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"account_status","arguments":{}}}'

A successful response contains your organisation's credit balance and key metadata in result._meta.precursor and in the parsed result.content[0].text.

4. Use the tools

Once connected, your agent can name any tool from the tools index directly. For example:

Pull intelligence on CVE-2024-3094 and tell me which vendors are currently affected, plus any KEV status.

The agent will pick get_cve (and likely get_shadowserver_cve_activity) on its own. You don't need to specify the tool name; just describe the question. The descriptions in tools/list are written to be selected correctly by the model.