A Free Competitive-Intelligence MCP Server (No Account, No API Key)

If you already do competitor research inside an AI assistant, the friction is the tab switch: you ask Claude or ChatGPT about a rival, it answers from whatever it happens to know, and then you go and check. An MCP server removes the switch. It gives the assistant a tool it can call mid-conversation, with the result coming back into the same thread.

Dozier runs one such server, and it is free — no account, no API key, no email. This post is the plain description: what the server does, how to add it to Claude, Claude Code, Cursor, or ChatGPT, what a result looks like, and exactly where the free scope ends. It is deliberately short on promises. The server exposes a single tool, and that tool does one thing.

What the server does

The server exposes one tool, run_exposure_audit(domain). You give it a bare company domain — acme.com, not a URL — and it runs Dozier’s Exposure Audit on that company: a public-source read on how the company is currently represented and, by extension, how AI answer engines are likely to describe it. The result is a short headline read, the top findings, and the number of public sources behind them.

That is the same audit you can run from the free AI visibility check on the website. The MCP server is a second door onto it, not a different product: same research, same caps, same cache. A domain that was audited recently comes back instantly; a fresh one takes a couple of minutes.

Two mechanics are worth knowing before you wire it up, because they are not what a tool call usually looks like:

• It is two calls, not one.The first call for a new domain starts the research in the background and returns a “started” message. Call the tool again with the same domain in two to three minutes and it returns the finished result. Most assistants will do this for you if you say “check again.”
• It is capped.Free usage is limited per visitor and site-wide each day so the server stays up for everyone. When a cap is hit, the tool says so in plain text — it does not hang or return a half result.

Every completed result ends with a link to view the same result on dozier.io. That page shows the identical findings, can be forwarded to a colleague, and has the free check on it so they can run their own.

How to add it

The transport is MCP Streamable HTTP with no authentication, and the endpoint is the same for every client:

https://dozier.io/backend/mcp

Claude

In claude.ai, go to Settings → Connectors → Add custom connector, and paste the endpoint URL above. No authentication step follows.

Step-by-step with screenshots: the Claude setup guide.

Claude Code

Run this in a terminal to add the Dozier MCP server:

claude mcp add --transport http dozier https://dozier.io/backend/mcp

Cursor and other MCP clients

Add this entry to your client's MCP server config:

{
  "mcpServers": {
    "dozier": {
      "url": "https://dozier.io/backend/mcp",
      "type": "http"
    }
  }
}

ChatGPT

Requires a paid ChatGPT plan (Plus and up). In ChatGPT, go to Settings → Apps → Advanced settings and turn on Developer mode, then add a connector with no authentication and this server URL:

https://dozier.io/backend/mcp

The OpenAI Responses API can also call it as a remote MCP tool; the exact JSON is on the MCP server page, alongside per-client setup guides.

What a result looks like

Ask your assistant something like “run an exposure audit on acme.com” and, once the second call lands, you get a short markdown block: the domain, a one-paragraph read of how the company currently comes across, up to five findings with a line of detail each, a source count and date, and the link to the result on dozier.io. It reads as plain text in any client. There are no charts and no scores.

What you do with it is the useful part. Run it on your own domain first — the gap between how you describe yourself and how public sources describe you is usually the first thing worth fixing. Then run it on two or three rivals and ask the assistant to compare. That comparison is a decent first draft of a positioning note, and it cost nothing.

Where the free scope ends

Being clear about the edges is the whole point of a post like this:

• Companies only.The tool takes a domain and nothing else. Names, IP addresses, and single-word inputs are rejected, and that rule is enforced on the server — not something an assistant can talk its way around.
• A snapshot, not a watch. The audit is a point-in-time read from public sources. It does not keep looking, and it will not tell you when something changes.
• Summary depth. You get the headline and top findings, not the full cited Dossier with an evidence trail behind every line.
• One tool.There is no tracked-rival list, no alerting, and no account-scoped data behind this server. It reads nothing of yours — no files, no browsing history, no workspace.
• Capped.Per-visitor and site-wide daily limits, stated in the tool’s own response when reached.

The continuous version — rivals tracked on a schedule, cited Dossiers, and Radaralerts when a rival’s pricing, messaging, or hiring actually changes — is the paid product, and it is not behind this server. Authenticated MCP access to a workspace is a separate thing we may build later; if it ships, it will be announced as such rather than quietly widening a free tool.

Why it is free

The honest answer is distribution. “What do AI answer engines say about you” is the question Dozier is built around, and an MCP server is the cheapest way to let someone ask it from the place they already work. If the answer is useful, the link at the end of the result is where the rest of the product lives. If it is not, you have spent nothing — and you are welcome to keep the server installed anyway. For the wider set of free options, the roundup of free competitive intelligence tools covers the field; for using a general assistant on its own, see ChatGPT for competitor research.