Tools & Resources

BrowserTools MCP exposes 16 tools, 3 prompts and 5 resource templates. Every tool declares an MCP output schema, so your client receives structured data rather than prose it has to parse. Read-only tools are annotated readOnlyHint, which lets clients auto-approve them safely. Every tool that reads from a page accepts an optional tabId; the log tools also accept allTabs. See working across several tabs.

Reading the page

Tool	What it does	Notable arguments
`getConsoleLogs`	Console output from the current tab	`keywords`, `limit`, `offset`
`getConsoleErrors`	Error-level output and uncaught exceptions	`keywords`, `limit`, `offset`
`getNetworkLogs`	XHR and fetch requests with status and timing	`urlKeywords`, `bodyKeywords`, `limit`, `offset`
`getNetworkErrors`	Only failed and 4xx/5xx requests	`urlKeywords`, `bodyKeywords`, `limit`, `offset`
`getSelectedElement`	The element selected in the Elements panel	`tabId`
`getPageInfo`	Which page the browser is currently on	`tabId`
`getBrowserStorage`	localStorage, sessionStorage and cookies	`kinds`, `includeValues`

Results are returned newest-first and always report total alongside returned, so an agent can tell when it is only seeing part of the picture. When a read has to be cut short, the tool attaches a resource_link to the full history rather than silently dropping the rest.

getBrowserStorage withholds values by default — it returns the key names and a count. Pass includeValues: true to read them. Cookie access additionally requires the optional permission you grant from the BrowserTools panel.

Acting on the browser

Tool	What it does	Notable arguments
`takeScreenshot`	Screenshot returned as an image, plus its file path	`name`, `tabId`
`refreshBrowser`	Reloads the inspected tab	`tabId`
`wipeLogs`	Clears captured telemetry before a clean reproduction	`tabId`

takeScreenshot returns path, name, mimeType, bytes, imageIncluded, tabId and url. Check url if you are not certain which page was captured.

Screenshots are held to a byte budget (3 MB by default). A capture that would exceed it is re-encoded as JPEG and, if still too large, downscaled — a dense viewport on a high-DPI display can otherwise run past 13 MB, which is more than a model’s context can take. If an image still cannot fit, it is written to disk and the tool returns the path with imageIncluded: false; read it back through the screenshot resource.

Orientation

Tool	What it does
`getConnectionStatus`	Whether the extension is connected, and capture counts
`listBrowserTabs`	Every tab with DevTools open, and the id to address it by

Audits

Tool	What it does
`runPerformanceAudit`	Lighthouse performance audit with Core Web Vitals
`runAccessibilityAudit`	Contrast, labels, semantics and screen-reader support
`runSEOAudit`	Metadata, indexability and crawlability
`runBestPracticesAudit`	Security, deprecated APIs and modern-web hygiene

All four take an optional url and default to the page currently open in the browser. Each launches a separate browser and takes up to a minute. Reports are flat and consistently shaped across all four categories:

{
  "category": "accessibility",
  "score": 72,                       // 0-100, or null
  "metadata": { "url": "...", "timestamp": "...", "lighthouseVersion": "13.4.1" },
  "summary": { "failed": 4, "passed": 1, "manual": 1, "informative": 1, "notApplicable": 1 },
  "issues": [                        // failing audits, heaviest first
    {
      "id": "color-contrast",
      "impact": "critical",          // critical | serious | moderate | minor
      "details": { "items": [...], "omittedItems": 0 }
    }
  ],
  "metrics": { }                     // Core Web Vitals, performance only
}

omittedItems reports how many detail rows were withheld to keep the payload small. Critical issues are never truncated, and the unabridged report is always available as a resource.

Prompts

Guidance ships as MCP prompts rather than tools, so it costs nothing until you ask for it. In 1.x these were tools that returned static text on every request, which spent context whether you wanted it or not. Pick them from your client’s prompt menu — in Cursor and Claude Code, type /.

Prompt	What it does
`debuggerMode`	A systematic reproduce-then-diagnose loop
`auditMode`	A full audit pass turned into a prioritised work list
`nextjsSeoAudit`	An SEO audit with Next.js-specific checks

Resources

Whole-history payloads are exposed as MCP resources rather than inlined, and tools link to them with resource_link so your agent fetches them only when it decides to.

Resource	What it is
`browser-tools://console/{tabId\|all}`	Every console entry, with no per-call size budget
`browser-tools://network/{tabId\|all}`	Every captured request, including bodies
`browser-tools://har/{tabId\|all}`	The same traffic as a HAR 1.2 file
`browser-tools://screenshot/{name}`	A screenshot you captured earlier, by the name in its result
`browser-tools://audit/{reportId}`	The unabridged Lighthouse result, by the `reportId` in its summary

Log tools attach a link when a read had to be cut short; network reads always offer the HAR; screenshots always link to the stored image, which is the only way to see one too large to inline. The 20 most recent full audit reports are kept under audits/ in the screenshot directory.

The HAR resource opens directly in Chrome’s Network panel and in most HTTP tooling, so you can hand a captured session to something outside your agent.

Exposing fewer tools

Sixteen tools is a lot of surface for a client that only needs a couple. Trim it with --only or --exclude:

npx @agentdeskai/browser-tools-mcp --only getConsoleErrors,getNetworkErrors,takeScreenshot

See configuration for the full list of flags.

Get Started

Reference

Upgrading & Security

Project

Reading the page

Acting on the browser

Orientation

Audits

Prompts

Resources

Exposing fewer tools

​Reading the page

​Acting on the browser

​Orientation

​Audits

​Prompts

​Resources

​Exposing fewer tools

Reading the page

Acting on the browser

Orientation

Audits

Prompts

Resources

Exposing fewer tools