MCP Server for the gym

[ollmer]

MCP Server that wraps up the BrowserGym gym, could be configured for any subset of task actions.
Short docs:

python browsergym/core/src/browsergym/utils/mcp_server.py -h

usage: python browsergym/core/src/browsergym/utils/mcp_server.py [options]

BrowserGym MCP server

options:
  -h, --help            show this help message and exit
  -t TASK_ID, --task_id TASK_ID
                        Task ID to run (default: browsergym/openended)
  -l, --headless        Run in headless mode (default: False)
  -r RECORD_VIDEO_DIR, --record_video_dir RECORD_VIDEO_DIR
                        Directory to save recorded videos (default: None)
  --demo_mode {off,default,all_blue,only_visible_elements}
                        Demo mode for action set (default: off)
  --timeout_ms TIMEOUT_MS
                        Timeout in milliseconds for each step (default: 10000)
  --subset {chat,infeas,bid,coord,nav,tab,miniwob_all,miniwob_shi17,miniwob_liu18,miniwob_humphreys22,webarena,visualwebarena,workarena,workarena++,weblinx,assistantbench}
                        Subset of actions to use (default: workarena++)
  --validate_actions VALIDATE_ACTIONS [VALIDATE_ACTIONS ...]
                        Names of actions for which validation should be performed (default: ['click', 'goto'])

To run Dev UI: mcp dev browsergym/core/src/browsergym/utils/mcp_server.py -e browsergym/core/

How to debug and run tool calls manually with web UI:

1. run mcp dev browsergym/core/src/browsergym/utils/mcp_server.py -e browsergym/core/
2. open http://127.0.0.1:6274 in browser
3. click Connect on the left panel.

How to run only the server with a visible browser:

python browsergym/core/src/browsergym/utils/mcp_server.py

Description by Korbit AI

What change is being made?

Introduce an MCP server for BrowserGym, enhancing its functionality to support remote procedural calls and facilitating environment interaction through a command-line interface, with necessary modifications to the project's configuration files to accommodate this feature.

Why are these changes being made?

These changes are being made to enable BrowserGym to operate as a server, handling web automation tasks remotely via MCP, which aligns with the goals of increasing the framework's flexibility and interoperability in automation environments. The addition of a CLI option enhances user experience by allowing more granular control over task execution parameters.

Is this description stale? Ask me to generate a new description by commenting /korbit-generate-pr-description

[korbit-ai]

Based on your review schedule, I'll hold off on reviewing this PR until it's marked as ready for review. If you'd like me to take a look now, comment /korbit-review.

Your admin can change your review schedule in the Korbit Console

[korbit-ai]

Review by Korbit AI

Korbit automatically attempts to detect when you fix issues in new commits.

Category	Issue	Status
	Missing Return Type Hints ▹ view	✅ Fix detected
	Critical Method Missing Type Hints ▹ view	✅ Fix detected
	Direct Access to Internal Implementation ▹ view	✅ Fix detected
	Redundant Dictionary Operations ▹ view	✅ Fix detected
	Over-broad exception handling ▹ view
	Global Configuration State ▹ view	✅ Fix detected

Files scanned

​

File Path	Reviewed
browsergym/core/src/browsergym/utils/mcp_server.py	✅
browsergym/core/src/browsergym/core/env.py	✅

Explore our documentation to understand the languages and file types we support and the files we ignore.

Need a new review? Comment /korbit-review on this PR and I'll review your latest changes.

Korbit Guide: Usage and Customization

Interacting with Korbit

• You can manually ask Korbit to review your PR using the /korbit-review command in a comment at the root of your PR.
• You can ask Korbit to generate a new PR description using the /korbit-generate-pr-description command in any comment on your PR.
• Too many Korbit comments? I can resolve all my comment threads if you use the /korbit-resolve command in any comment on your PR.
• On any given comment that Korbit raises on your pull request, you can have a discussion with Korbit by replying to the comment.
• Help train Korbit to improve your reviews by giving a 👍 or 👎 on the comments Korbit posts.

Customizing Korbit

• Check out our docs on how you can make Korbit work best for you and your team.
• Customize Korbit for your organization through the Korbit Console.

Current Korbit Configuration

General Settings

​

Setting	Value
Review Schedule	Automatic excluding drafts
Max Issue Count	10
Automatic PR Descriptions	✅

Issue Categories

​

Category	Enabled
Documentation	✅
Logging	✅
Error Handling	✅
Readability	✅
Design	✅
Performance	✅
Security	✅
Functionality	✅

Feedback and Support

• Tell us what you think of Korbit
• Schedule a call with our team
• Email us @ support@korbit.ai

[recursix]

Very cool. Let's dicuss further what's left to do for broader integration i.e what are the current limitations, how can it be used in AgentLab (probably need to automatically spawn a server per job).

[recursix]

All clean. Shouldn't break anything.