Skip to content

Competitive analysis with Claude and Cursor

How do I get Claude, Claude Code or Cursor to do competitor research?

Connect your AI client to the PMM Tools MCP server. PMM Tools captures each competitor page as clean markdown and a screenshot, with its source URL and capture date, and your agent reads those source files directly instead of a summary of them. Claude Code, Claude Desktop and Cursor connect today with an API key. ChatGPT can't connect yet, because its connectors require OAuth, which this server does not support yet.

How it works

  1. 1. Track your competitors

    Create a competitor from its domain. PMM Tools finds pages through the sitemap and site navigation, and captures each one as markdown and a screenshot with its source URL and capture date.

  2. 2. Connect your agent

    Generate an API key in PMM Tools and add the MCP server to your client, with the key sent as a bearer token. The command below is the Claude Code one; Claude Desktop and Cursor have their own setup guides.

  3. 3. Ask in plain language

    Your agent picks the tools it needs: listing competitors, reading captured pages, adding or excluding URLs, and starting refreshes. The analysis happens in your agent, on the source pages.

Connect Claude Code

Generate a key in PMM Tools, then run this with your key in place of the placeholder.

claude mcp add pmm --transport http --scope user \
  'https://www.pmmtools.com/api/mcp' \
  --header 'Authorization: Bearer pmm_live_YOUR_KEY'

Setup guides: Claude Code, Claude Desktop, Cursor.

Prompts to try

  • What competitors am I tracking in PMM Tools, and when was each one last captured?

    Tools: pmm_competitor_list

  • Read the captured pricing and features pages for [competitor] and summarize how their packaging differs from ours. Cite the capture date for each point.

    Tools: pmm_competitor_canonical_snapshots

  • Add [URL] to [competitor], start a refresh, and tell me when the run finishes.

    Tools: pmm_competitor_add_urls, pmm_competitor_run_now, pmm_run_status

  • Go through [competitor]'s captured pages and exclude any that aren't useful for positioning work, with a reason for each.

    Tools: pmm_competitor_get_urls, pmm_competitor_validate

Questions

Does my agent get summaries or the pages themselves?

The pages. Each tracked URL is kept as markdown and a screenshot with its source URL and capture date. Your agent does the analysis, so you can check any claim it makes against the captured page.

Which AI clients can connect?

Claude Code, Claude Desktop and Cursor, each with an API key sent as a bearer token; the MCP quickstart has a guide for each. ChatGPT can't connect yet: its connectors require OAuth, and OAuth token verification on this server is not live.

What can the agent do with the MCP server?

The server has 26 tools. They cover listing competitors, adding and excluding URLs, reading captured pages, scheduling and starting runs, and checking a run's status. The MCP quickstart lists every tool.

Is MCP access included in every plan?

Yes. Every paid plan includes MCP access, with a budget of 30 to 120 requests per hour depending on the plan. Reading captured pages does not use snapshot credits; starting a capture does.

See pricingConnect your agent