Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
You can use Browser Use with Codex in two distinct ways: connect a Model Context Protocol (MCP) server, or use Browser Use’s CLI/skill workflow, which explicitly lists Codex as a supported coding-agent environment. The hosted MCP server delegates a task to a browser in Browser Use’s cloud; the local stdio MCP server exposes browser controls for a local browser. OpenAI documents how Codex generally connects to MCP servers, but Browser Use’s published MCP examples do not provide a Codex-specific configuration block. So treat the hosted endpoint and authentication details below as vendor-documented, and verify that your current Codex version can send the required header before attempting a direct connection.
What Browser Use MCP does with Codex
MCP, or Model Context Protocol, lets a client discover and call tools exposed by a server. In this pairing, Codex is the client and Browser Use supplies browser-automation tools. Browser Use documents a hosted HTTP MCP endpoint at https://api.browser-use.com/mcp; its integration reference says HTTP clients generally need streamable HTTP and a request header for authentication. The documented header name is x-browser-use-api-key.
That describes the Browser Use service and its client examples, not a verified Codex setup. OpenAI’s Codex documentation shows a general remote-server workflow using codex mcp add, codex mcp list, or a server entry in ~/.codex/config.toml. It does not establish that a particular Browser Use endpoint, header syntax, or authentication method works in every current Codex release. See OpenAI’s Docs MCP guide and Browser Use’s integration reference for the respective vendor details.
Choose a connection method
| Method | Where the browser runs | How the agent works | What to know |
|---|---|---|---|
| Browser Use hosted MCP | Browser Use cloud | Codex calls MCP tools; Browser Use describes its service as handing a whole task to an agent in a cloud browser and returning the result. | Hosted endpoint and API-key header are documented for named clients. Codex-specific header configuration is not shown in Browser Use’s MCP page. |
| Browser Use local stdio MCP | A local browser environment | The MCP server exposes direct browser-control tools. | Browser Use documents a local launch command and environment variables, including an LLM API key. Requirements depend on the selected model and current package version. |
| Browser Use CLI/skill for coding agents | Local Chrome, Browser Use cloud, or a CDP endpoint | A coding-agent workflow uses the Browser Use CLI/skill rather than simply adding the hosted MCP endpoint. | Browser Use explicitly lists Codex for this workflow. Follow its current installation instructions at the coding-agent page. |
Choose hosted MCP when you want to delegate browser work to Browser Use’s cloud service through MCP tools. Choose local MCP when the goal is direct control of a local browser through stdio. Choose the CLI/skill route if you want Browser Use’s documented Codex-oriented coding-agent setup, or need to select among local Chrome, cloud browser, and a CDP endpoint. The available sources do not establish that these approaches have identical session context, privacy properties, compatibility, or reliability.
#1 Best Overall
Connect Codex to the hosted Browser Use MCP server
The two vendors document complementary pieces: Browser Use publishes its endpoint and header for several named clients; OpenAI publishes Codex’s general MCP configuration mechanisms. Neither source gives a tested, copy-and-paste Browser Use configuration for Codex. Use the following as a verification workflow, not as a claim that a guessed command is vendor-approved.
- Check your Codex version and current MCP options. In a terminal, run
codex --helpand inspect the current OpenAI MCP guide. Confirm whether your installed CLI supports remote HTTP MCP servers and how it attaches custom headers. Codex CLI and IDE extension share MCP configuration according to OpenAI’s Docs MCP guide. - Confirm Browser Use account access and key. Use Browser Use’s current MCP page and integration documentation to establish what account or API key is required. Do not paste a secret key into a public repository, prompt, or shared config file.
- Verify the endpoint and authentication header. Browser Use lists
https://api.browser-use.com/mcpand the header namex-browser-use-api-key. Check the live documentation for current endpoint behavior, exact key issuance steps, and whether Codex’s current configuration accepts the header in the required form. - Add the server only using syntax confirmed by current Codex documentation. OpenAI documents the general CLI pattern
codex mcp add ... --url ...and a~/.codex/config.tomlserver entry. Do not paste a speculative header flag or TOML field: the available Browser Use examples are for Claude Desktop, Cursor, and Windsurf, not Codex. If Codex cannot attach the required header to a remote server, use the documented Browser Use CLI/skill route instead. - Check what Codex registered. OpenAI’s documented CLI check is
codex mcp list. Confirm the server appears and that any configured secret is stored safely. - Test a small, low-risk browser task. Ask Codex to use Browser Use to open a public page and report a simple visible fact. Confirm it invokes the expected tool and returns a result. Do not use a sensitive logged-in session until you have reviewed where that session and task data go.
Browser Use’s hosted MCP page names six tools: run_session, get_session, send_task, stop_session, get_session_messages, and list_sessions. These names describe Browser Use’s hosted service; tool availability and behavior in Codex depend on a successful connection and the current service implementation. See Browser Use’s MCP page.
Run Browser Use’s local stdio MCP server
Browser Use’s integration reference gives this launch command for its local MCP server:
uvx --from 'browser-use[cli]' browser-use --mcp
This starts the server process over stdio; it is not the same as pointing Codex at the hosted HTTP endpoint. The reference also shows environment variables for an LLM API key. Configure the variables required by the current Browser Use version and the model you choose. Do not assume that local browser control removes model-provider requirements or that the command alone is a complete Codex configuration.
- Install or make available
uvxaccording to its own current instructions. - Check Browser Use’s integration reference for the current package extras, required environment variables, and tool setup.
- Configure Codex’s local stdio MCP server using the current syntax in OpenAI’s MCP guide. That guide confirms Codex’s general configuration support; it does not supply a Browser Use-specific local-server entry.
- Restart or reload the Codex client as needed, inspect its MCP server list, and test with a non-sensitive page.
Use this route when you specifically want Browser Use’s direct local browser controls. Local execution does not, by itself, establish that every part of the workflow or model inference stays on your machine; check the selected model and current Browser Use documentation.
Use Browser Use’s CLI/skill workflow for Codex
Browser Use explicitly names Codex on its coding-agent page and describes installing its CLI/skill, then connecting it to one of three browser modes: local Chrome, a Browser Use cloud browser, or any CDP endpoint. This is a separate documented route from adding https://api.browser-use.com/mcp as a hosted MCP server. Follow the current steps on Browser Use’s coding-agent page; the exact install and connection commands can change, so use the vendor’s current instructions rather than substituting a command inferred from MCP examples.
Rank #3
Pick the browser mode based on the task: local Chrome when the workflow calls for the local browser; cloud when you want the Browser Use cloud browser; or a CDP endpoint when you already have a compatible browser endpoint. The available documentation names those choices but does not warrant broader claims about privacy, anti-bot success, or compatibility with every browser setup.
Cost and practical limits
Browser Use’s undated hosted MCP page listed a price of $0.02 per browser-hour, plus model tokens, and a $15 signup credit for new signups with no card. These are vendor-listed figures, not independent measurements; pricing and the signup offer may change. Check the current pricing information before budgeting or creating an account. The cited price applies to the hosted service; it does not establish the total cost of local MCP or CLI/skill use, where model-provider charges or other dependencies may apply.
The same page reports an 82% solve rate across 106 internal tasks and 17¢ per solved task on its hardest internal set. Those are Browser Use’s self-reported internal benchmark figures, not an independent comparison or a promise about your task. They are not a reliable substitute for testing your own workflow.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Troubleshoot a connection or browser task
- Codex does not list the server: Check the exact configuration file and syntax for your installed Codex version, then run
codex mcp list. For an stdio server, make sure the launch command is available to Codex and that its process can start. - The hosted server rejects authentication: Re-check the API key and the required
x-browser-use-api-keyheader against Browser Use’s current integration reference. A correct URL alone is not sufficient if the service requires the header. - You cannot find a Codex header field or option: Do not guess a flag or TOML property based on another client’s example. Confirm current Codex support in OpenAI’s guide; use Browser Use’s explicitly Codex-listed CLI/skill path if the hosted MCP header cannot be configured.
- The stdio server exits immediately: Check that
uvxis installed and on the process path, that the package can be resolved, and that required model/API environment variables are present. Consult the current integration reference for version-specific dependencies. - The tool connects but cannot complete the task: Separate connection success from task success. Make the request smaller, specify the page and desired result clearly, and inspect available session messages with the hosted tools where applicable. The sources do not guarantee success on a particular site or challenge.
- The result is stale or the browser is in the wrong state: Start a fresh session if the selected workflow allows it, and explicitly state which page state or interaction is needed. Do not assume hosted, local, and CDP modes share cookies or session state.
- A sensitive page is involved: Stop and verify the browser location, authentication context, and service/model data handling before sending credentials or private content. The cited setup pages do not establish a universal privacy guarantee.
Or skip the browser setup
If your actual goal is a static website screenshot rather than having Codex interact with a live browser, ScreenshotNeo is a separate website screenshot API and MCP server for developers. It cannot replace browser task automation, but it can return a screenshot or PDF from one GET request. The API supports PNG, JPEG, or WebP output and many capture controls; see the ScreenshotNeo API documentation.
Example cURL request, using the documented API pattern and a sample target URL:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
curl -G "https://api.screenshotneo.com/v1/shot" -d access_key=YOUR_API_KEY --data-urlencode url=https://stripe.com -o shot.webp
Cookie/consent banners, newsletter popups, and chat widgets are removed before capture; each cleanup step can be turned off. Bot checks, blank pages, timeouts, failed loads, and cache hits cost nothing, and responses identify page verdict and billing status in headers. Its MCP server offers take_screenshot, get_page_info, and capture_pdf for AI agents. The Free plan includes 1,000 screenshots per month with no card; paid plans start at $5 for 3,000 screenshots.
Sign up for ScreenshotNeo’s free plan to try 1,000 screenshots a month with no card.
Frequently Asked Questions
Does Browser Use publish a Codex-specific hosted MCP configuration?
The Browser Use MCP material lists its endpoint and authentication header for several named clients, but does not show a Codex-specific configuration block. OpenAI documents Codex MCP setup generally; verify current header support before configuring the hosted endpoint.
Free tools Windows power users keep installed
One-click scans. No signup required.
Is Browser Use CLI/skill the same as its MCP server?
No. Browser Use documents the CLI/skill as a coding-agent workflow that explicitly includes Codex and supports local Chrome, cloud browser, or CDP modes. The hosted MCP and local stdio MCP are separate server connection methods.
Can ScreenshotNeo operate a browser for Codex?
No. ScreenshotNeo captures website screenshots and PDFs; it is an alternative for static capture tasks, not interactive browser automation.
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.
