hanzili

comet-mcp

Community hanzili
Updated

MCP Server connecting to Perplexity Comet browser

comet-mcp

Give Claude Code a browser that thinks.

An MCP server that connects Claude Code to Perplexity Comet - enabling agentic web browsing, deep research, and real-time task monitoring.

Why?

Claude Code is powerful, but it can't browse the web autonomously. Comet can browse, research, and execute multi-step web tasks - but it's isolated in a browser.

comet-mcp bridges them. Now Claude can:

  • Delegate web research to Comet's agentic browser
  • Monitor browsing progress in real-time
  • Intervene if tasks go off track
  • Get comprehensive research reports

Quick Start

1. Configure Claude Code

Add to ~/.claude.json (scope: user) or .mcp.json (scope: project):

{
  "mcpServers": {
    "comet-bridge": {
      "command": "npx",
      "args": ["-y", "comet-mcp"]
    }
  }
}

Or install globally first:

npm install -g comet-mcp

Then use:

{
  "mcpServers": {
    "comet-bridge": {
      "command": "comet-mcp"
    }
  }
}

2. Start Comet Browser

Download Perplexity Comet and launch with remote debugging:

# macOS
/Applications/Comet.app/Contents/MacOS/Comet --remote-debugging-port=9222

3. Use in Claude Code

You: "Use Comet to research the top AI frameworks in 2025"
Claude: [connects to Comet, delegates research, monitors progress, returns results]

Tools (6)

Tool Description
comet_connect Connect to Comet (auto-starts if needed)
comet_ask Send a task to Comet (non-blocking)
comet_poll Check task progress and get results
comet_stop Stop current task if off-track
comet_screenshot Capture current page
comet_mode Switch modes: search, research, labs, learn

Modes

Mode Best For
search Quick web searches
research Deep, comprehensive analysis
labs Data visualization & analytics
learn Educational explanations

Prompting Tips

When sending tasks to Comet via comet_ask, focus on context and goals rather than step-by-step instructions:

Good prompts:

"I need to find the best React state management libraries for a large-scale app.
Compare their bundle sizes, learning curves, and community support."

"Help me understand how Stripe's pricing API works. I'm building a SaaS
and need to implement usage-based billing."

Less effective prompts:

"Go to google.com, search for React state management, click the first result..."

Claude understands the task context; Comet handles the browsing. Let them collaborate - Comet will figure out the best way to navigate and gather information.

Example Workflow

# Claude Code conversation

You: Find the top 3 trending Python repos on GitHub

Claude: I'll use Comet to browse GitHub directly.

[comet_connect] → Connected to Perplexity
[comet_mode research] → Switched to research mode
[comet_ask "Find top 3 trending Python repos on GitHub today with their star counts"] → Task sent
[comet_poll] → Status: WORKING, Navigating to github.com/trending...
[comet_poll] → Status: WORKING, Filtering by Python...
[comet_poll] → Status: COMPLETED

Results:
1. project-a - 2.3k stars today
2. project-b - 1.8k stars today
3. project-c - 1.2k stars today

Architecture

Claude Code ←→ MCP Protocol ←→ comet-mcp ←→ Chrome DevTools Protocol ←→ Comet Browser
                                                                              ↓
                                                                    Perplexity AI (agentic browsing)

Requirements

Troubleshooting

"Cannot connect to Comet"

  • Make sure Comet is running with --remote-debugging-port=9222
  • Check if port 9222 is available

"Tools not showing in Claude"

  • Restart Claude Code after config changes
  • Verify ~/.mcp.json syntax

License

MIT

Built for the Claude Code community.

Report Issues · Contribute

MCP Server · Populars

MCP Server · New

    Dictation354

    Paper Fetch Skill

    Fetch papers as agent-ready markdown — DOI/URL/title in, structured full text out. CLI · MCP · Skill.

    Community Dictation354
    Kaelio

    ktx

    ktx is the context layer for analytics agents

    Community Kaelio
    Goldentrii

    AgentRecall

    Persistent, correction-driven memory for AI agents. Cross-session, cross-platform (Claude Code, Codex, Gemini — any MCP client). Learns from mistakes, compresses context to save tokens, consolidates knowledge overnight. npm: agent-recall-mcp

    Community Goldentrii
    aws-solutions-library-samples

    ABAP Accelerator MCP Server

    ABAP Accelerator is an MCP server that helps organizations create, test, document, and transform SAP ABAP code faster and with higher code accuracy.

    howardpen9

    kimi-code-mcp

    MCP server for Claude Code × Kimi K2.5 (256K context) — delegate bulk codebase analysis to Kimi, save 90% on token costs. Session caching, parallel agents, TypeScript.

    Community howardpen9