Introduction

Forgeline is a small MCP server you run on your own machine. It exposes a handful of tools — file I/O, shell execution, browser control, URL fetching — to any MCP-compatible client: Claude, ChatGPT, Notion AI, or your own.

Who it's for

Developers who already talk to an AI all day and want that conversation to reach into their actual environment without shipping their data to yet another SaaS. Forgeline runs on 127.0.0.1 by default; if you want remote access, you wrap it in a Cloudflare tunnel with bearer auth.

How it's structured

• Server: a single Python binary, one config file. No database.
• Workspace root: the only directory the server is allowed to read or write in.
• Tools: named operations the model can invoke — e.g. read_file, run_command, cdp_attach.
• Bridge: optional Cloudflare tunnel that maps localhost to an HTTPS URL so remote clients can reach the server.

Minimal example

Read a file from your workspace via any MCP client:

{
  "method": "tools/call",
  "params": {
    "name": "read_file",
    "arguments": { "path": "notes/inbox.md" }
  }
}

The server enforces path safety — attempts to escape the workspace root return an error:

{ "error": "path_outside_workspace", "path": "/etc/passwd" }

Continue to Quickstart to get a working server running in under a minute.