Spin up an agent and toolkit at the Room level for quick testing. The Room, agent, tools, and chatbot will shut down automatically when the Room is closed:

​
Environment Variables Needed

  • HYPERBROWSER_API_KEY: <app or user token>

​
1. Install Meshagent CLI and Tool Images

pip install "meshagent[all]"

​
2. Sign up and Authenticate

Follow these instructions to create an account and log in.

​
3. Launch the Hyperbrowser MCP Service Inside a Test Room

meshagent service test --room=test \
  --role=agent \
  --image=meshagent/mcp-hyperbrowser:latest \
  --env MESHAGENT_PORT=8001 --env HYPERBROWSER_API_KEY=<your-token> \
  --port="num=8001 path=/webhook liveness=/ type=meshagent.callable" \
  --name=mcp-hyperbrowser-service-test
  • This command starts a Meshagent Room named test with a Hyperbrowser MCP agent available to all Room participants.
  • Rooms auto-close if inactive; tools/agents are removed when the Room closes.

​
4. Start a Chatbot Agent in the Same Room

meshagent chatbot join --room=test \
  --agent-name=mcp-hyperbrowser \
  --name=mcp-hyperbrowser \
  --toolkit=mcp-hyperbrowser
  • This launches a chatbot able to use the Hyperbrowser toolkit in the Room.
  • Multiple toolkits can be used by the agent in the same Room.
  • The CLI will provide a link to the Room in the output.

​
5. Interact in the Browser

Visit the Room link provided, and send a message to the agent to use Hyperbrowser tools inside the Room!

​
Project Level Deployment: Persistent Service for All Rooms

For production, deploy persistent MCP and chatbot services at the project level—agents and tools will automatically register to every new Room upon its creation. This removes the need to run the MCP server or chatbot locally or manage Room-by-Room deployments.

​
1. Create the Persistent Hyperbrowser MCP Service

meshagent service create \
  --role=agent \
  --image=meshagent/mcp-hyperbrowser:latest \
  --env MESHAGENT_PORT=8001 --env HYPERBROWSER_API_KEY=<your-token> \
  --port="num=8001 path=/webhook liveness=/ type=meshagent.callable" \
  --name=mcp-hyperbrowser-service

​
2. Create a Persistent Chatbot Service

meshagent service create \
  --image="meshagent/cli:latest" \
  --port="num=9001 path=/agent liveness=/ type=meshagent.callable participant_name=mcp-hyperbrowser-chatbot" \
  --env="MESHAGENT_PORT=9001" \
  --name="mcp-hyperbrowser-chatbot-service" \
  --command="meshagent chatbot service --agent-name=mcp-hyperbrowser-chatbot --toolkit=mcp-hyperbrowser"
Now every Room in your Meshagent Project will automatically include these MCP tools and chatbot!

​
Tools Available

These tools are accessible in Meshagent Rooms via the MCP Hyperbrowser toolkit:
ToolDescription
browser_use_agentOpen-source browser automation agent for fast, efficient, cost-effective browser tasks using a cloud browser.
claude_computer_use_agentUses Anthropic Claude to autonomously execute complex browser tasks with advanced reasoning using a cloud browser.
crawl_webpagesCrawl a website from a starting URL and explore linked pages.
create_profileCreates a new persistent Hyperbrowser profile.
delete_profileDeletes an existing persistent Hyperbrowser profile by profileId.
extract_structured_dataExtract structured data from webpages using a schema and prompt.
list_profilesList existing persistent Hyperbrowser profiles.
openai_computer_use_agentUses OpenAI to autonomously execute general-purpose browser-based tasks using a cloud browser.
scrape_webpageScrape and extract webpage content in various formats.
search_with_bingSearch the web using Bing.

​
Tool Details

​
browser_use_agent

Cloud browser automation agent for explicitly defined, fast, and cost-effective browser interactions. Best for highly detailed, step-by-step actions and straightforward tasks. Params:
  • task (string): What to perform in the browser (REQUIRED)
  • maxSteps (integer, optional)
  • returnStepInfo (boolean, optional, default: false)
  • sessionOptions (object, optional)

​
claude_computer_use_agent

Invoke Anthropic Claude to handle complex web interactions needing context-sensitive reasoning. Params:
  • task (string): What to perform in the browser (REQUIRED)
  • maxSteps (integer, optional)
  • returnStepInfo (boolean, optional, default: false)
  • sessionOptions (object, optional)

​
crawl_webpages

Crawl a site from a URL, optionally following links. Useful for data collection, site mapping, or large-scale scraping. Params:
  • followLinks (boolean, REQUIRED)
  • outputFormat (array, REQUIRED)
  • url (string, REQUIRED)
  • ignoreSitemap (boolean, optional)
  • maxPages (integer, optional)
  • sessionOptions (object, optional)

​
create_profile

Create a new persistent profile.

​
delete_profile

Delete a specified persistent profile. Params:
  • profileId (string, REQUIRED)

​
extract_structured_data

Extract structured data from webpages matching a schema, e.g., via prompt and JSON schema. Params:
  • prompt (string, REQUIRED)
  • urls (array, REQUIRED)
  • schema (string, optional)
  • sessionOptions (object, optional)

​
list_profiles

List your persistent profiles, with pagination. Params:
  • limit (integer, optional)
  • page (integer, optional)

​
openai_computer_use_agent

Run browser tasks using OpenAI’s models for general-purpose automation. Params:
  • task (string, REQUIRED)
  • maxSteps (integer, optional)
  • returnStepInfo (boolean, optional, default: false)
  • sessionOptions (object, optional)

​
scrape_webpage

Extract raw content, html, or screenshots from a single webpage. Params:
  • outputFormat (array, REQUIRED)
  • url (string, REQUIRED)
  • sessionOptions (object, optional)

​
search_with_bing

Web search using Bing. Params:
  • query (string, REQUIRED)
  • numResults (integer, optional)
  • sessionOptions (object, optional)
Use the Hyperbrowser MCP server via Meshagent for secure, zero-ops, and scalable browser automation tools in every collaboration Room or project!