mantis

Name: mantis
Author: perzianhq

Local

github.com/perzianhqBrowser Automation2Updated 2d ago

Repository

About mantis

mantis is an MCP server that lets AI agents browse pages through concise, numbered text using a Playwright-based interface. It is designed for agents handling web tasks, returning changes between steps rather than relying on screenshots.

Key features

01Numbered text view of web pages
02Playwright-based browser interface
03Incremental page changes between steps
04Support for multi-step web workflows

Use cases

01Browse websites without screenshot-based input
02Complete long, multi-step browser tasks
03Give agents a compact representation of page changes

MCP support

mantis runs as an MCP server for agents such as Claude Code, Codex, and Cursor. It exposes its browser interface through MCP.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config

{
  "mcpServers": {
    "perzianhq-mantis": {
      "command": "npx",
      "args": [
        "-y",
        "mantis-browser"
      ]
    }
  }
}

Run commandnpx -y mantis-browser

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages

npmmantis-browser0.1.10stdio

Tools

Saved tool list · updated 1d ago

Parameter	Type	Required	Description
url	string	Yes

Details

From the project README

Lets AI agents browse the web as short, numbered text instead of screenshots.

perzian.dev/mantis · by Perzian

It works with Claude Code, Codex, Cursor or your own agent as an MCP server. Install in two commands.

How it works

Agent sends open/click to Mantis; Mantis reads the page in Chromium, numbers each control and returns only what changed; CAPTCHA, 2FA and payment go to you

What the agent sees

A shop page with a cookie popup becomes 8 numbered lines; after click #3 only the changes come back

Results

Same tasks, same pass criteria, three ways to browse. Mantis vs the accessibility tree (Playwright MCP) vs screenshots.

Long multi-step workflows (live-long-fixtures: 16 local tasks with in-page hazards, 3 runs per approach, code-graded). Score is out of 48; time is mean agent execution per run.

Model		Mantis	Accessibility tree	Screenshots
`gpt-6-luna`	Correct	48/48	48/48	42/48
	Time	103s	132s	163s
	Input tokens	17.3M	28.0M	31.0M
	Tool calls	1190	1685	2774
`claude-haiku-5-5`	Correct	48/48	48/48	48/48
	Time	42s	59s	64s
	Input tokens	12.1M	18.7M	27.1M
	Tool calls	1501	1827	2876

Real sites (live-v1 v1.7: 30 tasks, 3 runs per approach, judged blind). Success out of 90:

Model	Mantis	Accessibility tree	Screenshots
`gpt-6-luna`	89	90	85
`claude-haiku-5-5`	87	86	89

Net: accuracy is on par with the tree (within 1 run of 90 on real sites, equal on long workflows), at roughly 30% fewer tokens and calls on long tasks.

Where Mantis is better

Long workflows. Same accuracy as the tree, with 22–29% less time, 35–38% fewer input tokens and 18–29% fewer tool calls, because each step returns only what changed.
Versus screenshots. Fewer tokens and calls everywhere; Luna also missed 6 of 48 where Mantis and the tree did not.
Short tasks. Roughly a tie with the tree. The edge grows with task length.

Method and caveats: bench/RESULTS.md.

Learn more

How it works in detail
Claude skill: teaches Claude how to read Mantis output
Install, configuration and tool reference
Re-running the benchmarks

License

MIT

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

Read the getting-started guide

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.