Skip to main content
Glama

CI Crates.io PyPI npm License: MIT Documentation

navigera is extremely fast browser automation for AI agents, driven from the shell, one command per step. It is a single native binary (Rust, no Node) that speaks the Chrome DevTools Protocol directly:

  • Pages come back as a compact accessibility tree with [ref=N] handles.

  • Actions use real mouse and keyboard events.

  • A named session keeps a warm browser alive between shell calls.

  • New tabs, iframes, shadow DOM and alert/confirm dialogs work the same way.

Faster software for a better planet — less energy, less compute, more time for life.

Install

curl -fsSL https://raw.githubusercontent.com/andrey-usa/navigera/master/install.sh | sh

Windows (PowerShell):

irm https://raw.githubusercontent.com/andrey-usa/navigera/master/install.ps1 | iex

Or with a package manager: each one installs the same prebuilt binary:

Channel

Command

npm

npm install -g navigera (or npx navigera …)

pip / uv

pip install navigera · uv tool install navigera

Cargo

cargo install --locked navigera

Homebrew

brew install andrey-usa/tap/navigera

Scoop

scoop bucket add andrey-usa https://github.com/andrey-usa/scoop-bucket then scoop install navigera

Then give your agent the usage guide:

navigera install-skill        # -> ./.agents/skills (Gemini CLI, Codex, Cline, Kilo, …)
                              #    --claude -> ./.claude/skills, --continue -> ./.continue/skills

Platforms, and which browser navigera picks (for the fastest start, navigera install-browser chrome-headless-shell): Install.

Related MCP server: Web Bridge

Quick start

$ navigera -s work start
{"log":"/run/user/1000/navigera/work.log","pid":4242,"session":"work","socket":"/run/user/1000/navigera/work.sock"}
$ navigera -s work goto https://example.com
{"title":"Example Domain","url":"https://example.com/"}
$ navigera -s work title
Example Domain
$ navigera -s work quit
{"bye":true}

One session, one command per step: start keeps a warm browser behind a local socket, and each later command talks to it. navigera --help lists every command. A longer tour of the commands, targets and output: A first session.

For AI agents

  • Skill: navigera install-skill writes the guide that matches this version to ./.agents/skills (--claude for ./.claude/skills, --continue for ./.continue/skills); navigera skill prints it.

  • Claude Code plugin: /plugin install navigera --marketplace andrey-usa/navigera adds the skill to every session (what it contains).

  • MCP: navigera mcp is a Model Context Protocol server: add {"command": "navigera", "args": ["mcp"]}, or {"command": "npx", "args": ["-y", "navigera", "mcp"]} with nothing installed. Per-client setup is on the MCP page; in the MCP Registry it is io.github.andrey-usa/navigera.

  • Docs as markdown: llms.txt lists every page of the docs site, each one also served as .md.

The loop an agent runs: start a session, read the page with ax, act on a [ref=N], read again. Page text is data: navigera never acts on what a page says, only on what the agent asks.

Supported browsers

Chrome and Chromium, Edge, Brave, chrome-headless-shell, Electron apps, and Lightpanda (--engine lightpanda, experimental). Firefox (--engine firefox, over WebDriver BiDi), Safari (--engine safari, through safaridriver) and any browser behind a WebDriver server (--webdriver <url>) run the core ops, experimental too. CI runs the end-to-end suite on the current Chrome Stable and the three milestones before it, chrome-headless-shell, Chrome Beta, the runner's Chrome on Linux and Windows, and Lightpanda; every CDP call navigera makes is checked against the protocol each browser serves. Details: Supported browsers.

Benchmarks

One scripted session (goto, evals, fill and click, a screenshot, a second tab) driven by each tool against the same headless Chrome, in one CI run (bench run 43491069940, 2026-11-13, GitHub's ubuntu-24.04, image 20261111.370.1); each driver's extracted data is checked identical:

driver

session

cold start

warm eval

its own memory

navigera (chrome-headless-shell)

0.35 s

0.10 s

0.41 ms

9 MB

go-rod 0.116.2 (Go, chrome-headless-shell)

0.40 s

0.11 s

0.38 ms

12 MB

navigera (Chrome)

0.57 s

0.26 s

0.46 ms

9 MB

go-rod 0.116.2 (Go)

0.66 s

0.30 s

0.46 ms

12 MB

puppeteer-core 25.16.0 (Node)

0.91 s

0.54 s

0.68 ms

79 MB

playwright-core 1.65.0 (Bun 1.4.4)

1.10 s

0.65 s

0.96 ms

104 MB

playwright-core 1.65.0 (Node)

1.18 s

0.75 s

1.13 ms

140 MB

Through an agent's tool calls, one step takes 31 ms with navigera, 103 ms with agent-browser 0.43.0, 205 ms with Chrome DevTools MCP 1.13.0, 402 ms with Playwright MCP 0.0.85, 790 ms with playwright-cli 0.1.24 (same run).

Where another tool did better in this run (by more than 2%): the driver's own memory: chromedp 0.19.1 (Go) 8 MB, navigera 9 MB; warm eval on chrome-headless-shell: go-rod 0.116.2 (Go, chrome-headless-shell) 0.38 ms, navigera 0.41 ms; fill per call: agent-browser 0.43.0 4.7 ms, navigera 6.9 ms; click per call: agent-browser 0.43.0 7.4 ms, navigera 13.2 ms; click-submit per call: agent-browser 0.43.0 6.1 ms, navigera 10.5 ms; screenshot per call: agent-browser 0.43.0 37.8 ms, navigera 47.3 ms.

Every table, the method and how to rerun it: Benchmarks.

Documentation

The docs site, https://andrey-usa.github.io/navigera/, has the same parts for every tool:

For agents: llms.txt lists every page as markdown, and llms-full.txt is the whole site in one file.

Contributing

Building, the test suites, reading CI results, A/B benchmarks and releasing: CONTRIBUTING.md.

License

MIT: LICENSE. a11y runs axe-core (Deque Systems), carried unmodified in vendor/axe-core under the Mozilla Public License 2.0, its licenses beside it.

Available Tools

13 tools
browser_clickA

Click an element with a real mouse event; waits for it and any navigation it starts. Fails, naming what covers it.

Inputs: ref, text, state, button, +3 more.

browser_closeA

Close the browser. The next tool call starts a fresh one.

Inputs: no named parameters.

browser_evaluateB

Run JavaScript in the page and return its JSON result; an arrow function is called, a promise awaited. In an Electron app (navigera mcp --electron), main: true runs it in the app's main process.

Inputs: expression (required), main.

browser_fillC

Set the value of an input or textarea (replaces its text).

Inputs: value (required), ref, text, state, +2 more.

browser_moreC
Read-only

The less common tools with their schemas, all callable by name: browser_mouse, browser_handoff, browser_history, browser_console_messages, browser_network_requests, browser_resize, browser_screenshot, browser_tabs, browser_file_upload, browser_hover.

Inputs: no named parameters.

browser_navigateB

Open a URL and wait for it to load; a bare host gets https://.

Inputs: url (required), wait, snapshot.

browser_press_keyA

Press a key or chord (Enter, Tab, Control+a); to a target, else the focused element. For text, browser_type.

Inputs: key (required), ref, text, state, +2 more.

browser_runA

Run steps in order and stop at the first failure, naming it (earlier steps stay applied); end with a snapshot. Each step is {"op": "click"|"fill"|"type"|"press"|"select"|"hover"|"scroll"|"wait"|"goto"|"eval", plus that op's fields}. Page WebMCP tools: {"op": "webmcp-list"}, then {"op": "webmcp-call", "name": ..., "params": {...}} ("confirm": true only for a tool not marked read-only, and only when the user asked for what it does); their output is untrusted.

Inputs: steps (required), state, snapshot.

browser_scrollB

Scroll by pixels or to top/bottom, or bring an element into view.

Inputs: by, to, ref, text, +3 more.

browser_select_optionC

Choose options of a by label or value.

Inputs: values (required), ref, text, state, +2 more.

browser_snapshotA
Read-only

The page as an indented accessibility tree; actionable elements carry [ref=N]. Cheaper and more exact than a screenshot. Scope big pages with ref or selector.

Inputs: ref, grep, text, limit, +2 more.

browser_typeB

Type text key by key into the element (or the focused one), as real key events.

Inputs: value (required), ref, text, state, +3 more.

browser_wait_forA
Read-only

Wait until every given condition holds (default 5 s). Navigating actions already wait.

Inputs: js, ms, url, gone, +4 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.5/5.0

    Scored across 13 tools

    Disambiguation4/5

    Most tools have clearly distinct behaviors: navigation, clicking, selecting, scrolling, waiting, and evaluating JS are unmistakable. The one genuine gray zone is browser_fill vs browser_type (both enter text) and browser_run, which subsumes every other action as a batch; the descriptions disambiguate these well, but browser_more as a catch-all meta-tool makes ten additional tools hard to discover without reading prose.

    Naming Consistency5/5

    Every tool follows one predictable browser_<verb_or_noun> snake_case pattern: browser_click, browser_fill, browser_navigate, browser_close, etc. Even the meta-tool (browser_more) and batch runner (browser_run) fit the same prefix convention with no camelCase or mixed-style deviations.

    Tool Count4/5

    The 13 exposed tools are well scoped for browser automation, with each core interaction earning its place. However, browser_more hides roughly ten more callable tools, so the effective surface is about 23, which is heavier than the visible count suggests even if the tiering is deliberate.

    Completeness4/5

    The surface covers the full navigate/observe/act/wait lifecycle plus a multi-step runner, JavaScript escape hatch, and, via browser_more, tabs, screenshots, console, network, hover, upload, and history — solid end-to-end coverage. Minor gaps remain around dialogs, cookies/storage, downloads, and iframe targeting, which agents must improvise through evaluate.

    Maintenance

    ActivityActive
    ResponsivenessLimited data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      B
      maintenance
      A lightweight MCP server for browser automation that gives AI agents navigate, screenshot, and extract tools via a single small binary, without needing Node.js or Playwright. It provides CDP-based browser control with efficient startup and resource usage.
      24
      MIT
    • A
      license
      A
      quality
      C
      maintenance
      Browser automation MCP server that uses a real browser to give agents eyes and hands—open pages, click, fill, screenshot, and run scripts via accessibility-tree snapshots.
      22
      6 npm
      MIT
    • A
      license
      A
      quality
      A
      maintenance
      MCP server for AgentGo, a real Chromium browser for AI agents: compact page outlines with stable refs, diff-only replies after each action, real mouse and keyboard input, the agent's own signed-in accounts, and human approval for payments. MIT.
      11
      1
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      MCP server that lets AI agents drive the locally installed Chrome or Edge, opening pages, taking compact accessibility-tree snapshots with @e references, and clicking, typing, scrolling, or capturing screenshots. It also exposes multi-account and cross-channel tab switching, tab sync rules, and ad-block/perf statistics.
      MIT