Harnesses

Wire Workbooks into your coding tool.

Workbooks ships an MCP (Model Context Protocol) server inside the CLI. Any AI coding tool that speaks MCP can use it to scaffold projects, build workbooks, publish them to Studio, and manage runtime config — as structured tool calls, no copy-paste.

Prerequisite. Install the CLI first: npm install -g @work.books/cli. The MCP server runs as workbook mcp serve and is what every harness on this page connects to.

Pick your harness

Claude Code Official Workbooks skill + MCP. Best support. →

OpenClaw Peter Steinberger's open-source agent (350k+ stars on GitHub). MCP via ~/.openclaw/config.json. → Cursor MCP via .cursor/mcp.json (project) or user settings. →

OpenAI Codex CLI agent from OpenAI. MCP via ~/.codex/config.toml. → Cline VS Code extension. MCP through the Cline settings UI. → Continue Open-source assistant for VS Code and JetBrains. → Windsurf Codeium's AI-native IDE. MCP via ~/.codeium/windsurf/mcp_config.json. → Hermes Agent Nous Research's self-improving agent with persistent memory. MCP under [tools.mcp] in ~/.hermes/config.toml. → Pi Minimal terminal coding agent CLI. MCP via ~/.pi/agent/config.json. → More tools Roo Code, Goose, OpenHands, Trae, GitHub Copilot, Replit Agent, Lovable, v0 — one install prompt per tool. →

What an agent can do once it's wired up

list_groups — see every group across your workspace
init_project — scaffold a new workbook at a path
build — compile a project to one HTML file
publish — upload a built workbook to Studio
inspect — read metadata, permissions, source-bundle stats from a built workbook
unbundle — recover source from a built workbook
get_runtime_config / set_runtime_config — manage WB_ENV values

Don't see your harness?

Any tool that speaks MCP can talk to workbook mcp serve. If yours isn't listed, the generic stdio config under Agent setup works. Open an issue on GitHub and we'll add a dedicated page.