MCP Comet

Turn Perplexity Comet into a production-grade MCP research engine

7 modes · 13 tools · zero-friction setup · full browser control

Quick Start · Tool Reference · Architecture · Issues

Why MCP Comet
Demo
Quick Start
Research Modes
Toolset at a Glance
Agent Workflows
CLI Power Ops
Architecture
Configuration
Compatibility
Contributing
License

Why MCP Comet

MCP Comet gives your agent more than a chat box. It gives your agent a complete research cockpit.

Run high-quality web research in Comet directly from MCP clients.
Switch between 7 purpose-built research modes instantly.
Pull sources, screenshots, conversations, tabs, and full page content.
Stay resilient with auto-connect, reconnect logic, and selector fallback strategies.

[!NOTE] If your workflow is "ask, verify, cite, and iterate", this is the server built for it.

Demo

MCP Comet deep-research demo

Quick Start

Prerequisites

Node.js >= 18
Perplexity Comet installed

[!TIP] Run mcp-comet detect to verify your setup before configuring.

Install

Option 1: Global install (recommended)

npm install -g @onestepat4time/mcp-comet

Option 2: Run without install

npx -y @onestepat4time/mcp-comet

Option 3: Local development

git clone https://github.com/OneStepAt4time/mcp-comet.git
cd mcp-comet
npm ci
npm run build

Configure

Add MCP Comet to your MCP client config.

Claude Desktop (~/.claude/claude_desktop_config.json):

{
  "mcpServers": {
    "mcp-comet": {
      "type": "stdio",
      "command": "mcp-comet",
      "args": ["start"]
    }
  }
}

Cursor (~/.cursor/mcp.json):

{
  "mcpServers": {
    "mcp-comet": {
      "type": "stdio",
      "command": "mcp-comet",
      "args": ["start"]
    }
  }
}

First Query

Prompt your agent with something like:

Use Comet in deep-research mode to analyze the global battery supply chain in 2026. Return a structured summary with all cited sources.

Your agent can chain comet_mode, comet_ask, comet_wait, and comet_get_sources automatically.

Research Modes

Choose the mode that matches the job.

Mode	Description	Example
`standard`	Fast factual lookups	"What is the latest CPI reading for Canada?"
`deep-research`	Multi-source investigations	"Map the 2026 AI chip supply chain and major risks."
`model-council`	Multi-perspective reasoning	"Debate arguments for and against UBI with tradeoffs."
`create`	Drafting and ideation	"Draft a technical explainer on WebAssembly in edge runtimes."
`learn`	Guided teaching	"Teach me B-trees step by step with examples."
`review`	Critical analysis	"Review this API design for security and reliability gaps."
`computer`	Browser-interactive tasks	"Open arXiv and find the newest papers on retrieval augmentation."

CLI example:

mcp-comet call comet_mode '{"mode":"deep-research"}'
mcp-comet call comet_ask '{"prompt":"Analyze current fusion startups by funding and milestones"}'
mcp-comet call comet_wait
mcp-comet call comet_get_sources

Toolset at a Glance

Session

Tool	Description
`comet_connect`	Connects to Comet or launches it
`comet_poll`	Returns live status and partial progress
`comet_wait`	Waits for completion and returns the full response
`comet_stop`	Stops a running task

Query

Tool	What It Does
`comet_ask`	Sends a prompt to Comet
`comet_mode`	Gets or switches active research mode

Content

Tool	What It Does
`comet_screenshot`	Captures PNG/JPEG screenshots
`comet_get_sources`	Extracts references, including collapsed citations
`comet_get_page_content`	Extracts page title and readable text

Navigation

Tool	What It Does
`comet_list_tabs`	Lists tabs by category
`comet_switch_tab`	Jumps to a tab by id or title
`comet_list_conversations`	Lists sidebar conversations
`comet_open_conversation`	Opens a specific conversation

Full reference: docs/tools.md

Agent Workflows

Goal	Flow
Deep research with citations	`comet_connect` -> `comet_mode(deep-research)` -> `comet_ask` -> `comet_wait` -> `comet_get_sources`
Multi-perspective debate	`comet_mode(model-council)` -> `comet_ask` -> `comet_wait`
Visual evidence capture	`comet_screenshot` -> pass image into your vision-capable model
Resume old investigations	`comet_list_conversations` -> `comet_open_conversation` -> `comet_get_page_content`

CLI Power Ops

Install globally:

npm install -g @onestepat4time/mcp-comet

Use directly:

# connectivity and diagnostics
mcp-comet detect
mcp-comet call comet_connect

# ask + wait pattern
mcp-comet call comet_ask '{"prompt":"What are the top AI safety papers this week?"}'
mcp-comet call comet_wait

# source extraction
mcp-comet call comet_get_sources

No install option:

npx -y @onestepat4time/mcp-comet

Architecture

MCP Tools
   -> UI Automation
      -> CDP Transport
         -> Perplexity Comet

Ordered selector strategies tolerate Comet UI changes.
Automatic version detection selects the correct selector set.
Auto-reconnect includes health checks with retry backoff.
Source extraction uses a second pass to expand collapsed citations.

Deep dive: docs/architecture.md

Configuration

Configuration Essentials

Most teams only tune these three:

Variable	Default	Change It When
`COMET_RESPONSE_TIMEOUT`	`180000`	`comet_wait` calls are timing out
`COMET_PATH`	auto-detect	Comet is in a non-standard install path
`COMET_LOG_LEVEL`	`info`	You need debug logs

Config file (mcp-comet.config.json) example:

{
  "responseTimeout": 300000,
  "logLevel": "debug"
}

More options and full env var reference: docs/configuration.md

Compatibility

Chrome Version	Selector Set	Status
145	v145	Supported

Unknown versions fall back to the latest known selector set.

Details and upgrade flow: docs/comet-compatibility.md

Contributing

PRs welcome. For non-trivial changes, please open an issue first to discuss what you'd like to change.

npm run lint && npm test

Full guide: docs/contributing.md

License

MIT

_{Built with dedication by OneStepAt4time}

@onestepat4time/mcp-comet

Quick Install