TIGER-AI-Lab / TIGER-AI-Lab/ClawBench
Harbor: register browser MCP tools for the default local runtime
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 795
- Forks
- 58
- Avg merge
- 2d 21h
- Merged PRs (30d)
- 18
Description
Problem
The default local-browser adapter emits CDP environment variables but no environment.mcp_servers. Kernel tasks already register pinned Playwright MCP. Harbor's stock Claude Code and Codex consume task MCP configuration to install browser tools; a CDP URL alone does not register those tools. This leaves the default browser mode without the same out-of-box tool configuration.
Scope / acceptance
- Emit the existing pinned Playwright MCP definition for both local and Kernel tasks, pointing at the existing recorded CDP session (
http://127.0.0.1:9223). - Preserve Kernel-only environment and lifecycle hooks.
- Load all V2 tasks in both modes with Harbor 0.22.0 and 0.23.0 and assert the MCP configuration survives Harbor parsing.
- Document that only agents consuming task MCP configuration are covered; do not claim every Harbor agent is end-to-end verified.
Child of #331; independent of #353 schema/staging hardening and #346 readiness changes. No new mandatory CI or image publication requirement.
@Perry2004 please review the default local-browser tool contract and the implementation PR linked below.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Locate the default local-browser adapter and the Kernel task MCP registration, then compare their task environment generation and Harbor parsing entry points. Load V2 tasks under Harbor 0.22.0 and 0.23.0 in both modes; done means the pinned Playwright MCP definition uses the recorded CDP session, Kernel-only hooks remain unchanged, and the supported agent scope is documented.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- devtools, testing-qa
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 66/100