Browser MCP

Connect a local MCP server and Chrome extension so an agent can use your existing browser profile. Pair the extension deliberately, then work in task tabs with the selected account.

Read one signed-in report

A read-only report lookup checks pairing and account context without changing business records. Use it to decide whether the existing Chrome session fits later tasks.

Report URL, Chrome profile, account, date range, fields, and expected units.

Copy the request into a new Floatboat conversation and provide the required material.

Download Floatboat

Your request for Floatboat

Preview and copy the request. It is not sent automatically from this page.

Reference: Browser MCP
https://github.com/BrowserMCP/mcp

Read the original documentation and check the current environment and dependencies. Propose any new installation or tool connection before proceeding.

My task:
Use Browser MCP with my paired test Chrome profile to read the specified report. Confirm the account and period, collect only approved fields, and do not alter settings or submit forms.

Expected deliverable:
A report summary with account context, selected period, and source pages. Include failed steps and unresolved questions when evidence is insufficient.

About Browser MCP

Each host-facing MCP process forwards calls to a per-user local daemon. The extension relays browser debugging operations only after the user compares a six-digit code and allows pairing. The host then works in task tabs backed by the selected profile; Stop and restricted-page rules remain enforced by the extension.

Choose Browser MCP when an existing Chromium profile and its signed-in pages matter more than provisioning a fresh cloud browser. The documented extension relay depends on Chromium debugging support; Firefox is not supported. The Chrome Web Store extension updates automatically, while a self-built copy requires rebuilding and reloading, with a compatible server version.

The repository uses Apache-2.0 and provides an MCP server and browser extension. It relies on the connected agent for task planning. Existing login state does not remove account permissions, and local browser execution does not determine how the host model processes page content.

How Browser MCP Fits the Browser Workflow

The browser connection has three layers: the host-facing MCP process, a local daemon, and the Chrome extension. Account state comes from the selected profile; the extension controls which actions can be relayed.

  • Local daemon and extension relay

    The host-facing MCP process forwards calls to a per-user daemon; the paired extension relays browser debugging operations over localhost.

  • Profile-backed task tabs

    Task tabs use the selected Chromium profile and its account state, rather than requiring every lookup to start from a fresh login.

  • Extension-side controls

    Pairing, Stop, restricted pages, and the separate incognito permission are enforced by the extension, not merely suggested to the agent.

Browser MCP Use Cases

Existing Chrome state is helpful for portal reports and profile-specific issues. The following tasks keep that continuity while separating draft preparation, read-only collection, and review of agent tab activity.

Read a profile-dependent portal report

Provide the report URL, account/workspace, date range, and columns. After pairing the intended Chrome profile, confirm those visible settings in a task tab and collect a sourced summary. If a total is absent, distinguish an empty period from a report that never finished loading.

Check a profile-dependent bug

When a bug occurs only in the existing profile, reproduce the specified interaction through the extension rather than switching immediately to a fresh browser. Record the relevant state and keep unrelated extensions or tabs out of the investigation until they are needed.

Prepare values in an authenticated form

Provide exact draft values and a clear stopping point. Have the agent fill the form in a task tab and report the visible validation state. The person reviewing the browser can then decide whether to submit; the extension connection does not grant blanket write approval.

Review task-tab activity

For a shared browsing session, ask the agent to name the pages it touched and summarize the observed result. Use the extension stop control if scope changes unexpectedly, then inspect the task tabs before resuming. A clean tool response cannot replace checking the actual pages.

Before Using Browser MCP

Prepare the intended Chrome profile, extension pairing, and client connection before granting browser access.

  • Chrome extension

    Install the official extension in the intended Chrome profile and keep its controls available.

  • MCP server and pairing

    Prepare the documented Node runtime, connect the MCP package, and approve the matching local pairing code.

Read One Report Through a Paired Chrome Profile

Supply a report URL, account/workspace name, date range, and named columns. The result should identify the selected account and period alongside the values, not merely confirm that Chrome opened.

  1. Connect the official components

    Install the official extension in the intended Chromium profile and configure the documented @browsermcp/mcp server in the host. Start a browser task and compare the six-digit code in the extension popup with the host before choosing Allow. If they differ, stop and inspect the selected server and profile.

  2. Establish the report context

    Open the supplied report in a task tab. Read the visible account/workspace and selected dates before collecting rows. Ask the user to resolve login or account selection when those values cannot be confirmed; an already signed-in profile may still contain the wrong account.

  3. Read and reconcile the requested fields

    Select the approved period and collect only the named columns. Return each value with the report URL and period, comparing a representative row or displayed total with the page. Distinguish an empty report from a loading error or incorrect filter.

  4. Resolve connection failures before repeating actions

    If control fails, check whether Stop is active, pairing is complete, and the page is permitted. Browser and extension pages are excluded; incognito requires a separate extension setting. Resume only after observing the intended page and profile again.

Related Software Projects

Compare the workflows, operating requirements, and source licenses of these projects.

Browser MCP FAQ

Is this Microsoft’s Playwright MCP?
No. Browser MCP is a separate project with its own MCP package, daemon, and Chrome extension. Both can automate pages, but their setup and browser-session model differ.
Must I install both parts?
The documented real-browser workflow uses the MCP server and Chrome extension together. A client configuration alone does not provide the extension or complete the pairing step.
Why is the extension installed but the browser unavailable?
Check the host server, intended profile, matching pairing codes, and Stop control. A self-built extension and a store copy enabled together may appear as two browsers; use the upstream build instructions to avoid that ambiguity.
Can it access incognito or internal pages?
The extension enforces its documented restrictions. Incognito requires an explicit browser setting, and internal browser/extension pages are restricted; do not promise arbitrary-page access.
Does it have an official Skill?
Use the official server, extension, and MCP documentation. The main repository does not document an official end-user Skill installation.
Does my model see the whole profile?
The agent receives the browser data exposed through its tool use. Scope the task and selected tabs carefully; using a local relay does not establish that observations stay away from a remote model.