Vessel Browser MCP
A Chromium browser built for agents with visible control: an MCP server, named sessions, checkpoints and approval prompts
- 139 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent drives a real browser: it opens pages, fills forms and uses saved sessions.
- Session files and the credential vault hold cookies, tokens and passwords.
Install
- Works in
Manual install
npm install -g @quanta-intellect/vessel-browserThe first npm command from the README; start with `vessel-browser`. The Linux alternative is the AppImage from GitHub Releases.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent drives a real browser: it opens pages, fills forms and uses saved sessions.
- Session files and the credential vault hold cookies, tokens and passwords.
This is third-party code. Review the repository files before installing.
What it does
Vessel is an open-source Chromium (Electron) browser built to work with agents. Pages open in an ordinary window, so you see where the agent goes and which actions it asks you to approve. An external MCP client such as Claude Code or Codex, as well as Hermes Agent and OpenClaw, drives it through a local MCP server on port 3100. A `vessel-browser-mcp --stdio` proxy supplies the token for clients automatically. Long tasks survive restarts. Named sessions save cookies and localStorage, checkpoints with notes give a return point, and `undo_last_action` rolls state back to just before the last mutating agent action. Significant actions are approved once, for a run or for a domain.
Who it is for. Developers of agent systems who need a browser with a visible interface and the ability to step into the agent work.
Good fit when
- An agent runs a long multi-source research and the work has to resume after a restart.
- Before submitting a form or taking another significant action, the agent must get your approval.
- A login is saved in a named session and reused in the next run.
- An agent is asked to watch a page for changes and show what changed.
Not a fit when
- You need headless automation on a server: Vessel is a windowed app.
- The work touches banking or other critical accounts: the authors themselves call the project early software.
- Paid features are needed for free: skills, the credential vault and screenshot analysis are part of Premium.
Example request
Open the pricing pages of three competitors in Vessel, save a checkpoint, compare prices and, before filling any form, show which fields you intend to fill.MCP
- Transport
- stdio, http
- Authentication
- API key
Limitations
The project is under active development, which matters for sensitive browsing and accounts. Linux (AppImage) is the main platform, Windows has an installer and the macOS package is built from source. Vessel must be running before an MCP client connects, and the token sits in `~/.config/vessel/mcp-auth.json`. Skills, the credential vault, screenshots with visual analysis and the changes timeline belong to paid Premium.
How to disable. Remove the vessel entry from the MCP client configuration and close the browser. Saved sessions live in the app data folder; delete them separately because they may contain cookies and tokens.
FAQ
How do I connect an agent?
Start Vessel and set the command `vessel-browser-mcp` with `--stdio` in the client configuration: the proxy reads the token from `~/.config/vessel/mcp-auth.json` itself. You can also connect directly at `http://127.0.0.1:3100/mcp` with the token supplied manually.
Where are saved sessions stored?
In the app data folder with restrictive file permissions. They may hold login cookies and tokens, so treat them as secrets.
Which models does the built-in chat support?
Anthropic, OpenAI, Ollama, llama.cpp, Mistral, xAI, Google Gemini, OpenRouter and any OpenAI-compatible endpoint.
What requires approval?
Sensitive agent actions. The approval card shows the action, redacted arguments and the domain. You approve once, for a run or for a domain, or reject with steering for the agent.
How do I undo the last agent action?
Through the Supervisor tab or the `undo_last_action` tool: the browser returns to the session snapshot taken before the last successful mutating action.
Discussion
No questions or reviews yet.
Related
MCP serversA separate browser for agents with a skill and MCP: Claude Code and Codex work in your logged-in accounts, and every session is recorded for replay
A real browser inside the terminal with a CLI for the agent: a coding agent opens sites and interacts with them in the same tab
MCP gives an agent a Playwright browser where a person can take over the same session and save an authentication profile
An MCP wrapper over the browser-use engine: a single tool drives a browser from a natural-language task.
A CLI and MCP server that lets an agent drive your real Chrome with its logged-in sessions: 103 commands across 36 platforms plus browser automation
An MCP bridge to control your browser from a coding agent: a Chrome extension and a set of tab and page automation tools
README in short
Install is possible as an AppImage, through npm or with the `install.sh` script that builds the app and creates the `vessel-browser-launch`, `-update` and `-status` helpers. The supervisor sidebar shows the run inbox, conversation threads, approvals and checkpoints, and bookmarks carry metadata for the agent.