Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

PinchTab

A single Go binary that runs Chrome as a local HTTP service for agents, with persistent profiles, text snapshots and site audits.

Reviewed by Sheela · Updated

View on GitHub Open source

Our rating

3.8/5

A lean local browser service that saves agents real tokens, from a young project whose security record so far argues for keeping it strictly on localhost.

What is PinchTab?

PinchTab is a standalone HTTP server, written in Go, that gives AI agents control of Chrome. It installs as a background daemon on localhost:9867, manages browser profiles and instances, and exposes navigation, accessibility-tree snapshots with stable element refs, clicks, form fills and text extraction through a CLI, an HTTP API with a bearer token, and an MCP server. The npm package also drops a pinchtab skill into Claude Code, Cursor, Windsurf, Codex and OpenClaw.

It runs headless or headed, keeps logged-in profiles across restarts, can run several isolated Chrome instances in parallel, and adds site audits and visual diffs for CI. An optional CloakBrowser provider targets sites that fingerprint stock Chromium.

The repo is MIT-licensed with about 10,500 GitHub stars; the latest release is v0.15.2 (26 August 2026). Seven security advisories were published between 5 and 22 March 2026, including a high-severity SSRF.

Is PinchTab worth it?

PinchTab's case is economy. Instead of feeding the model screenshots or raw HTML, it returns an accessibility snapshot with short element refs such as e5 that survive filtering and only expire on navigation, and a text mode the project puts at about 800 tokens a page against roughly 10,000 for the raw page. Its own benchmark against agent-browser, run on the Anthropic API, reports 9.5% to 20.3% lower cost, with the gap widening on longer 24-step tasks. Those are the maintainers' numbers, but the method and raw logs are published in the repo.

The architecture suits a developer's machine. One roughly 30 MB binary runs as a daemon, so every agent tool on the laptop shares one browser control plane, and named profiles keep cookies and logins so an agent can use your work account without a fresh sign-in each run. Headed mode, multiple parallel instances and attaching to a remote Chrome cover QA and testing jobs, and pinchtab audit adds screenshots, console errors, Core Web Vitals and a compare command that can fail a CI build on visual diffs.

The security defaults are cautious, and the record explains why they matter. It binds to 127.0.0.1, turns off sensitive endpoint families and attach mode, and restricts browsing to locally hosted sites until you widen the allowlist. Seven advisories were published in March 2026: SSRF through the download handler (rated high), token leakage in query strings, a rate limiter that was never applied, and a Windows command injection via profile names. Six list fixed versions; the Windows one lists none. The README is plain that the dashboard, API and MCP server are privileged surfaces not built for untrusted users or the public internet.

Against libraries such as Stagehand or browser-use, PinchTab is a service rather than a library, so it fits agents that already call tools over HTTP or MCP and do not want a Node or Python runtime embedded. Windows support is best-effort, and the release pace slowed after August, so it reads as a sharp local tool more than a shared automation platform.

Where it earns its place

  • A solo developer running Claude Code and Codex side by side

    Let both agents browse docs and dashboards without each launching its own browser

    Run pinchtab daemon install once; npm install -g pinchtab copies the skill into each agent's home, and both drive the same Chrome on localhost:9867. Free, and text snapshots keep page reads near 800 tokens.

  • A QA lead at a 15-person web agency

    Catch visual regressions before a client release

    Add pinchtab compare between production and staging with --fail-on-diff to CI, and pinchtab audit on the sitemap for console errors and Core Web Vitals. Runs in the Docker image bound to 127.0.0.1.

  • An operations analyst

    Have an agent download a weekly report from a logged-in portal

    Create a named headed profile, log in once by hand, and let the agent reuse that profile each week. The session cookies live on the machine, so treat the profile as a credential.

Before you commit

Setup
curl -fsSL https://pinchtab.com/install.sh | bash, brew install pinchtab/tap/pinchtab or npm install -g pinchtab, then pinchtab daemon install; pinchtab nav <url> --snap is the first command. Needs a local Chrome or Chromium; Docker image available.
Integrations
CLI, HTTP API with bearer token, MCP server, agent skills for Claude Code, Cursor, Windsurf, Codex and OpenClaw, plugins for Grok Build and OpenClaw; optional CloakBrowser provider.
Data and privacy
No telemetry or analytics, per the README; binds to 127.0.0.1 and profiles store browser sessions locally. MIT licence. Seven GitHub advisories (one high, six medium) published in March 2026; six have fixed versions.
Cost at scale
No paid edition. Cost is the machine running Chrome instances (the Docker example sets 2 GB of shared memory) plus model tokens, which the snapshot format is designed to reduce.
Limits
Primary testing is macOS and Linux; Windows is best-effort. Remote, container and distributed setups are operator-managed and must be secured by hand. Last release v0.15.2 in August 2026.

What’s good

  • Accessibility snapshots and text mode cut tokens per page
  • Single ~30 MB Go binary; daemon shared by every local agent
  • Persistent named profiles, headed or headless, parallel instances
  • Built-in site audit and visual compare for CI

What’s not

  • Seven security advisories in March 2026, one without a listed fix
  • Privileged API not designed for remote or multi-user exposure
  • Windows support is best-effort
  • Benchmarks are the maintainers' own; no release since August 2026

Best for

Developers who want one local, token-efficient Chrome service that all their coding agents and scripts can share.

PinchTab alternatives

PinchTab: common questions

Is PinchTab worth it?
A lean local browser service that saves agents real tokens, from a young project whose security record so far argues for keeping it strictly on localhost. We rate it 3.8 out of 5.
How much does PinchTab cost?
Free and open source (MIT).
What are the downsides of PinchTab?
Seven security advisories in March 2026, one without a listed fix; privileged API not designed for remote or multi-user exposure; windows support is best-effort; benchmarks are the maintainers' own; no release since August 2026.
Who is PinchTab best for?
Developers who want one local, token-efficient Chrome service that all their coding agents and scripts can share.
How much does PinchTab cost for a team or in production?
No paid edition. Cost is the machine running Chrome instances (the Docker example sets 2 GB of shared memory) plus model tokens, which the snapshot format is designed to reduce.
Is PinchTab safe for company data?
No telemetry or analytics, per the README; binds to 127.0.0.1 and profiles store browser sessions locally. MIT licence. Seven GitHub advisories (one high, six medium) published in March 2026; six have fixed versions.
What are the best PinchTab alternatives?
The closest alternatives we have reviewed are Browser Use, Stagehand, Skyvern and Magnitude.

Answers checked Oct 31, 2026.