What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
To use Agent Browser with GitHub pages, run the Agent Browser CLI as a stdio MCP server from an MCP-capable client, then ask that client to navigate, inspect, and interact with the page. The documented launch command is agent-browser with mcp as its argument. This is separate from GitHub’s repository-level MCP configuration for Copilot cloud agent and code review; the two setups serve different workflows.
What Agent Browser MCP does on GitHub
Agent Browser is a browser automation CLI for AI agents. Its documented browser engine uses Chrome or Chromium through CDP, and its MCP mode exposes browser functions to an MCP client over stdio. That lets a client work with a GitHub page as a browser user would: open a page, inspect its accessible content, and interact with visible controls.
The usual interaction loop relies on accessibility-tree snapshots. A snapshot gives the client a view of page content and compact element references such as @e1. The client can then act on a reference and take a new snapshot to inspect the changed page. References describe a particular current page state; after navigation or interaction, inspect a fresh snapshot rather than assuming an earlier reference is still valid.
This is useful for low-impact work such as opening a repository page, reading visible information, or locating a link. It is not the same thing as installing a GitHub integration, granting repository permissions, or configuring Copilot.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →#1 Best Overall
Agent Browser MCP and GitHub repository MCP are different setups
| Question | Agent Browser with a local MCP client | GitHub repository MCP for Copilot |
|---|---|---|
| Where is it configured? | In the MCP client’s configuration, using a client-specific file or settings screen. | In repository Settings under Copilot, then MCP servers. |
| What runs the work? | The MCP client launches the local agent-browser mcp server and uses browser tools. |
Copilot cloud agent and Copilot code review can use tools from the repository’s configured MCP servers. |
| What is the primary use? | Browser navigation and interaction with GitHub pages. | Repository-level MCP tools available to Copilot cloud agent and code review. |
| What MCP capabilities are documented? | Agent Browser provides browser tools; its core profile supplies everyday browser functions, with optional profiles for additional capabilities. | GitHub documents support for MCP tools, not MCP resources or prompts, in this workflow. |
| What should not be assumed? | The documented local stdio server is not established as directly supported in GitHub’s hosted repository setting. | GitHub’s built-in GitHub MCP server is not Agent Browser. |
GitHub’s documentation says repository-configured MCP tools can be used autonomously without an approval prompt. It recommends enabling only specific tools, especially read-only tools where practical. GitHub also says this configuration does not currently support remote MCP servers that use OAuth. Check GitHub’s repository MCP documentation for the current options and limits.
Connect Agent Browser to an MCP client
Install Agent Browser and the browser it depends on by following the project’s current installation guide. Exact installation commands and the MCP configuration-file location vary by operating system and MCP client, so use the project and client documentation rather than copying a configuration path intended for another client.
- Install the CLI and browser. Follow the current Agent Browser project guide for installation and browser setup. Confirm that the
agent-browsercommand is available in the environment from which your MCP client starts servers. - Open the MCP client’s server settings. Find that client’s instructions for adding a local stdio MCP server. The config file and JSON shape are client-specific.
- Set the server command and arguments. Configure the client to launch command
agent-browserwith the arguments array["mcp"]. In clients that use a JSON entry withcommandandargs, the relevant values are"agent-browser"and["mcp"]; do not assume that the enclosing keys or file path are universal. - Restart or reload the MCP client. Check the client’s server or tool list for Agent Browser’s browser tools. If it does not appear, check the client’s server logs and verify that its process can find the CLI and launch the browser.
- Start with a read-only page task. Ask the client to open a public GitHub repository page and report visible information, such as the repository name or the text of a visible link. Keep the first task read-only while you confirm the connection works.
The project’s core profile is intended for everyday browser functions. The documentation also lists optional profiles named network, state, debug, tabs, react, mobile, and all. Use the profile guidance in the project documentation to decide whether a task needs one; do not enable broader tools without a reason.
Use snapshots to inspect and navigate GitHub pages
In an MCP client, you normally ask the assistant to use the exposed browser tools; the client handles the tool calls. These equivalent CLI examples show the documented pattern using Agent Browser commands:
- Open a page:
agent-browser open https://github.com/vercel-labs/agent-browser - Inspect the page:
agent-browser snapshot - Use a reference from that snapshot: if the snapshot identifies a link or control as
@e1, an example interaction isagent-browser click @e1. - Inspect the new state: run
agent-browser snapshotagain after the click, navigation, or other page change.
In an MCP conversation, phrase the same task in terms of the outcome: open the repository page, inspect its accessible snapshot, identify a particular visible link, and report its destination or text. For actions that change repository state—such as submitting a form, creating an issue, or changing settings—be explicit about the intended change and required permissions. A page’s MCP metadata or a tool’s description does not grant authorization.
Keep browser access and GitHub permissions safe
Limit the task and the tools
Use only the browser capabilities needed for the task. Agent Browser’s project documentation cautions that page content and WebMCP metadata—including tool names, descriptions, schemas, annotations, and results—are untrusted website data. It states: “These labels are provenance cues, not a prompt-injection security boundary.” Discovery tells an agent what a page claims to offer; it does not make an action safe or authorized. Host permissions and your own approval process still govern tool execution. See the Agent Browser project documentation for its security guidance.
For GitHub’s repository-level Copilot MCP configuration, narrow the enabled tools to the minimum necessary. GitHub warns that Copilot cloud agent and code review may use configured MCP tools autonomously without asking for approval, so avoid enabling write-capable tools when a read-only task will do.
Protect login state
Agent Browser documents authentication persistence using profiles, sessions, state files, and an auth vault. Treat saved authentication data as a secret: the project README warns that state files contain session tokens in plaintext unless encryption at rest is configured. Do not commit state files to a repository, and use a trusted machine and a narrowly scoped session.
Best Value
Understand local debugging access
The project warns that a remote debugging port gives local processes full control of the browser. Run the browser in an environment you trust, avoid exposing debugging access unnecessarily, and do not assume that an MCP client’s tool list is a security boundary.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Troubleshooting connection and page problems
- The MCP client cannot find Agent Browser: The CLI may be installed in a shell-specific path that is not inherited by the MCP client. Verify the executable is available to the client process; if the client supports an absolute command path, use the installed path and reload the client.
- The server fails during startup: Check the client’s MCP server logs for command, argument, or browser-launch errors. Confirm the command is
agent-browserand the argument list containsmcp, then verify the browser installation using the project’s current setup instructions. - No browser tools appear: Reload the MCP client after changing its server configuration and inspect its status or logs. Confirm that the configuration was added to the MCP client—not to GitHub repository Settings, which is a separate Copilot configuration.
- A reference such as
@e1no longer works: The page may have changed since the snapshot. Take a new snapshot and use a reference from the current accessibility tree. - The page is blank, incomplete, or still loading: Wait for the page to settle and take another snapshot. A sign-in wall or other content not visible to the current browser session may prevent the information you requested from appearing.
- A GitHub action is rejected: Check whether the browser session is signed in and whether the account has permission for that action. Do not treat a visible control, discovered tool, or page-provided description as authorization.
- A session appears unexpectedly signed out: Check which profile or state the browser session is using. Protect saved session data and avoid sharing it in logs or source control.
Or skip the browser setup
If your goal is a screenshot of a GitHub page rather than interactive browser automation, ScreenshotNeo is a website screenshot API and MCP server. One GET request can return a PNG, JPEG, WebP, or PDF. It accepts cookie or consent banners like a visitor before capture and removes more than 60 known consent platforms, newsletter popups, and chat widgets; each of those steps can be turned off. Bot checks, blank pages, timeouts, failed loads, and cache hits are not billed, and responses include X-Page-Verdict and X-Billed headers. Its MCP server provides take_screenshot, get_page_info, and capture_pdf tools for Claude, Cursor, and other MCP clients.
For a public page, the one-call cURL example is:
curl -G "https://api.screenshotneo.com/v1/shot" -d access_key=YOUR_API_KEY --data-urlencode url=https://github.com/vercel-labs/agent-browser -o shot.webpSee the ScreenshotNeo API documentation for response formats and options. One thousand screenshots per month are free with no card; paid plans start at $5 for 3,000. Sign up for the free plan.
Frequently Asked Questions
Can Agent Browser click or submit GitHub controls through MCP?
Yes, browser interaction is the purpose of its tools, but a browser capability does not grant GitHub account permission. For any state-changing action, specify the intended change and confirm that the signed-in account is authorized.
Free tools Windows power users keep installed
One-click scans. No signup required.
Does GitHub’s repository MCP setting install Agent Browser?
No. It configures MCP servers for Copilot cloud agent and code review; Agent Browser’s documented setup is a local stdio server connected through an MCP client.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
