[Feature Request] Browser Cloud, Kane CLI and Rook integrations

Motivation

Disclosure up front: I work at TestMu AI (formerly LambdaTest). We build the three tools below, and I am raising this as an issue first, per CONTRIBUTING.md, rather than opening a PR.

Two gaps show up when a Workforce runs several browser agents and specialised workers in parallel.

The first is browser capacity. By default, Eigent's browser pool runs Chrome on the user's own machine, so parallel browser agents share one machine's CPU and memory. Eigent can attach to an existing CDP browser, but normalize_cdp_url keeps only scheme://host:port and the reachability check expects /json/version, so a hosted browser whose session or credentials live in the URL path or query cannot attach today.

The second is verification. Eigent shows which files changed in a session, with diffs and checkpoints, and Workforce agents plan, solve and verify together. What is not covered is grading a run against observed effects outside the workspace: the tools an agent called, the APIs it hit and the records it changed, compared with what it reported doing.

Solution

Three tools, usable independently and all opt-in, so Eigent's local-first default does not change.

Browser Cloud runs real Chrome sessions in the cloud for AI agents, with built-in parallelism, session persistence (cookies, local storage and login state), a tunnel to localhost or staging, and video, network and console logs for every session. A small change to how Eigent normalises CDP endpoints would let it attach to hosted sessions like these, and to any other hosted browser. https://www.testmuai.com/browser-cloud/

Kane CLI is an AI-driven testing tool for you and your agents. You describe a flow in natural language and it drives real Chrome from the terminal, returning a pass or fail with a replayable .evidence pack for every run. Its --agent flag switches output to structured NDJSON, so an agent can parse the result and decide what to do next. It also ships as an agent skill, which could fit Eigent's Skills. https://github.com/LambdaTest/kane-cli

Rook CLI grades what an agent actually changed rather than what it says it did. It reads your codebase, requirements docs or MCP servers, generates functional and adversarial scenarios, runs your agent against them, checks the real effects on tools, files and APIs, and reports Pass, Fail or Unable to Verify. Apache-2.0. https://www.testmuai.com/rook-cli/

We also run a hosted MCP server at https://mcp.lambdatest.com/mcp (Streamable HTTP, OAuth sign-in). Eigent's Custom remote option saves only a URL, so the route that should work today is a Custom local server running npx -y mcp-remote@latest https://mcp.lambdatest.com/mcp, the same pattern Eigent uses for Notion.

Alternatives

For browsers, the local browser pool, which works well while one machine has the capacity.

For verification, Workforce review plus the session diff view, which cover the workspace but not tool calls or external effects.

Additional context

Rook CLI is pre-alpha: commands and stored file formats can still change, though builds are published and the install is public. All three need a TestMu AI account.

Happy to shape the contribution however suits you: a skill, a small CDP change, docs only, or not at all. Wanted to ask before writing any code.