Forge MCP server

Forge is a helper for Claude Code that takes one big goal and breaks it into smaller pieces. It runs those pieces in parallel, checks each one, and retries the ones that fail. It is handy if you use Claude Code for tasks that touch many files and you keep losing track of what is happening.

What this MCP server does

You type a goal like /forge add audit logging for admin actions. Forge's planner looks at your code and proposes a plan made of modules with dependencies. You approve the plan. Then worker agents build each module in its own isolated copy of your project, running in parallel where possible. A reviewer checks the results, and a debugger steps in when something fails. At the end you get a summary of which modules finished and which did not.

Click to zoom

What you can do with it

• Break a large feature into a plan of modules with dependencies
• Run independent modules at the same time in separate worktrees
• Validate each module against commands and acceptance criteria you approved
• Retry failed modules with extra debugging context
• Resume an unfinished session instead of starting over
• Remember conventions and failure patterns across sessions
• Show live progress in your terminal status line

Try asking your AI

• “/forge add JWT auth with refresh tokens”
• “/forge build an audit log for admin actions”
• “/forge-status”
• “/forge-validate m2”

What it gives back to you

You get a proposed plan first, which you approve, modify, or abort. During the run you see progress messages like which module is executing, which passed, and which failed. At the end you get a short summary such as 4/4 modules completed. Forge also writes plans, logs, retry history, and memory into a local .forge folder in your project.

Before you start

Who it's for

Developers and technical teams who use Claude Code for multi-file features and want a structured plan, parallel execution, and validation.