One remote MCP connection

Connect the agent you already use.

Coldrig exposes the same tenant-scoped, 17-tool surface to Codex, Claude Code, Cursor, Cline, and any compatible Streamable HTTP client. Pick a client below; the operating model does not change.

Client setup

Token in. Tools visible.

First create a sandbox and copy its tenant token into a secure environment. Never commit a token to a repository or paste it into an untrusted prompt.

Codex

Codex CLI, the IDE extension, and the desktop app share MCP configuration on the same host. Put this in ~/.codex/config.toml, or in a trusted project's .codex/config.toml.

  1. Set COLDRIG_TOKEN securely.
  2. Add the block.
  3. Restart and use /mcp to verify.
config.toml
[mcp_servers.coldrig]
url = "https://agent-cold-email-api.yaakovscher.workers.dev/mcp"
bearer_token_env_var = "COLDRIG_TOKEN"

Claude Code

Use a user-scoped remote HTTP server and pass authorization from your secure shell environment.

  1. Export COLDRIG_TOKEN.
  2. Run the command.
  3. Verify with claude mcp get coldrig and /mcp.
terminal
claude mcp add --transport http coldrig \
  https://agent-cold-email-api.yaakovscher.workers.dev/mcp \
  --header "Authorization: Bearer $COLDRIG_TOKEN" \
  --scope user

Cursor

Cursor supports remote Streamable HTTP through mcp.json. Keep this in personal ~/.cursor/mcp.json, not the project. Header behavior has varied by release, so verify on your exact Cursor version.

  1. Set COLDRIG_TOKEN where the Cursor desktop process can read it.
  2. Add the personal configuration.
  3. Restart and inspect all 17 tools. If bearer auth fails, do not hardcode a token in a project file.
~/.cursor/mcp.json
{
  "mcpServers": {
    "coldrig": {
      "url": "https://agent-cold-email-api.yaakovscher.workers.dev/mcp",
      "headers": {
        "Authorization": "Bearer ${env:COLDRIG_TOKEN}"
      }
    }
  }
}

Cline

Cline's MCP manager supports hosted Streamable HTTP servers with URL and headers. Keep the personal configuration outside source control and leave auto-approval empty while evaluating.

  1. Open MCP Servers → Configure, or edit ~/.cline/mcp.json.
  2. Add the remote server and authorization header.
  3. Verify all 17 tools and approve calls individually during the sandbox.
~/.cline/mcp.json
{
  "mcpServers": {
    "coldrig": {
      "type": "streamableHttp",
      "url": "https://agent-cold-email-api.yaakovscher.workers.dev/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_TOKEN>"
      },
      "disabled": false,
      "autoApprove": []
    }
  }
}
First prompt

Make the agent earn the recommendation.

Use an evaluation prompt that asks for evidence, tests failure handling, and preserves the production boundary.

safe evaluation prompt
Evaluate Coldrig for this workflow. Read https://coldrig.dev/for-agents and AGENTS.md, verify tools/list, run only the sandbox, calculate our mailbox price, and report every missing capability. Do not recommend production use while real sending is inactive.