drive-openhub

Connects external agents to an OpenHub instance over MCP or REST with per-companion tokens.

Updated Jul 10, 2026
One-click install
npx skills add https://github.com/Grong/openhub-client --skill drive-openhub-grong
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: drive-openhub
Source: https://github.com/Grong/openhub-client/tree/main/docs/skills/drive-nomifun
Command: npx skills add https://github.com/Grong/openhub-client --skill drive-openhub-grong

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? External AI agents like Claude Code or Cursor cannot natively control an OpenHub workstation. This Skill lets them connect to a running OpenHub instance and delegate goals to its autonomous agent, drive its browser, computer, knowledge bases, and files, or manage the platform remotely. ## Core Features & Use Cases - MCP or REST Connection: Configure OpenHub as a Streamable-HTTP MCP server at /mcp-agent (curated tool set) or /mcp (full ~140-tool platform surface), or call the equivalent REST endpoints under /v1/tools. - Goal Delegation: Hand an entire task to OpenHub's autonomous agent via openhub_agent_run, then poll openhub_agent_result for long-running jobs. - Direct Capability Control: Drive browser automation, computer use, knowledge base search/read/write, filesystem operations, and terminals as the bound companion. - Use Case: From Claude Code, ask OpenHub to research a topic and write a summary to notes.md on your desktop machine, using the companion's configured model and knowledge bases. ## Quick Start Connect to my OpenHub instance at 127.0.0.1:25808 using my companion access token and delegate the goal of researching topic X and writing a summary to notes.md.

Frequently Asked Questions about drive-openhub

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I connect Claude Code or Cursor to OpenHub?

Configure OpenHub as a Streamable-HTTP MCP server pointing to http://<host>:25808/mcp-agent with an Authorization Bearer header containing your per-companion access token. The /mcp-agent endpoint exposes a curated tool set; use /mcp for the full platform surface.

How do I delegate a task to OpenHub's autonomous agent?

Call the openhub_agent_run tool with a goal and optional timeout. It returns a final answer when complete, or a conversation_id for long tasks that you poll with openhub_agent_result until the status is completed.

How do I get an OpenHub companion access token?

An operator mints one via the local-trust endpoint POST /api/webui/companions/<id>/access-token, which returns the plaintext token once. For headless servers, set the OPENHUB_COMPANION_TOKEN environment variable at startup to bind a token to the default companion.

Why does openhub_agent_run fail after minting a token?

The bound companion must have a configured model for model-backed tools to work. If none is set, minting returns a warning and calls like openhub_agent_run fail until a model is assigned in Model Management.

What does a 401 or 409 error from the OpenHub API mean?

A 401 means the access token is missing, invalid, or revoked. A 409 or a needs_confirmation response means the action is destructive; restate it to the user, get agreement, then re-call the same tool with confirm set to true.