Loading...
Everything you need to go from zero to your first build.
Prerequisite — Claude Code installed on your machine.
Download the ManyHands binary for your platform. No dependencies required — it's a single executable.
This detects your OS and architecture, downloads the binary to /usr/local/bin/manyhands, and makes it executable.
curl -fsSL https://api.manyhands.dev/cli/download/manyhands-darwin-arm64 -o ~/.local/bin/manyhands && chmod +x ~/.local/bin/manyhandscurl -fsSL https://api.manyhands.dev/cli/download/manyhands-linux-amd64 -o ~/.local/bin/manyhands && chmod +x ~/.local/bin/manyhandsLink your terminal to your ManyHands account. This opens your browser for a one-time sign-in.
Credentials are stored locally at ~/.manyhands/credentials.json and never leave your machine.
For API key or OAuth token users. If you use claude login (OAuth) you can skip this step.
Workers need a Claude credential to execute tasks. Choose one:
Your key stays on your machine — never sent to ManyHands servers.
Register the ManyHands MCP server so Claude can orchestrate builds on your behalf.
This tells Claude Code to launch manyhands mcp as a tool server. You only need to run this once.
Open Claude Code and describe what you want to build. Claude handles the rest.
Example — in Claude Code:
"Build me a real-time chat app with WebSocket support, user auth, and message persistence. Use FastAPI + React."
Claude will call plan_prompt to draft a structured spec, ask you clarifying questions, then call start_orchestration to open the configure page in your browser. Adjust workers, models, and budget — then hit Build.
Claude drafts a detailed project spec from your description and asks clarifying questions.
Review the build in your browser — pick models, worker count, budget cap, and QA phases.
Parallel AI workers research, architect, implement, and test your project on your machine.
These tools are available to Claude Code after you complete step 3.
plan_promptReturns a structured prompt template. Claude fills it in by asking you questions, then presents the finished spec for approval.
start_orchestrationKicks off a build run. Accepts an inline description or a file path, optional project name, and optional source_bundle for existing codebases. Opens the configure page in your browser.
bundle_projectPackages a local directory as a .tar.gz bundle. Excludes .git, node_modules, __pycache__, .venv, and files >50MB. Pass the result to start_orchestration as source_bundle.
check_statusCheck progress of a running build — completed tasks, active workers, current phase, and cost.
list_runsList your recent orchestration runs with status and metadata.
cancel_runCancel an in-progress build run. Kills local worker processes.
resume_runProResume a failed or canceled run from its latest snapshot. All completed work is restored. Creates a new run linked to the original.
iterate_runProContinue a completed run with new or changed requirements. The queen detects the scope change and runs delta research before generating new implementation tasks.
subscribeOpens the billing page in your browser to upgrade your plan. Required for resume_run and iterate_run.
You can also use the CLI directly without Claude Code.
manyhands login— Authenticate with your ManyHands accountmanyhands logout— Clear stored credentialsmanyhands start "Build a ..."— Start an orchestration run from a promptmanyhands status <run-id>— Check the status of a runmanyhands list— List recent runsmanyhands cancel <run-id>— Cancel an active runmanyhands mcp— Start the MCP server (used by Claude Code)