Proofsource

What Proofsource MCP server does

The Proofsource MCP server measures how often AI-generated answers mention or recommend a tracked brand. It works across ChatGPT, Perplexity, Claude, and Google AI Overviews, using buyer questions associated with the brand's tracked topics.

Results include mention rate, share of voice, average position, citation share, and sentiment. You can examine results by engine or by individual buyer question. Additional tools identify competitors named instead of the brand, the websites cited in answers, changes during a selected period, and gaps that may prevent AI systems from citing or recommending the brand.

A workspace can also record shipped fixes and later compare mention rates before and after the change. Verification classifies the result as moved, within normal answer variation, or awaiting enough samples.

How it works

Start by listing the brands in the workspace. Most brand-specific tools require the returned brand_id. Topics and buyer questions are available through list_topics; create_prompt can add a question to a topic for future engine runs, subject to the plan's question limit.

For a new measurement, add_brand accepts a website or an app store, Google Maps, YouTube, or Instagram link. Proofsource reads the supplied source, selects buyer questions, and begins an initial scan. get_scan_status reports progress by stage and engine. Once the scan is complete, visibility, competitor, citation, and change tools can retrieve the results.

The see_example tool provides an unsigned example using Tesla. It is intended to demonstrate the type of visibility data available; checking a user's own brand requires signing in.

Setup and configuration

The remote MCP endpoint is https://app.proofsource.co/v1/mcp. It uses OAuth 2.1 and a Proofsource email with a one-time sign-in code. In Claude, add the endpoint as a custom connector. In Claude Code, add it with the HTTP transport and then authenticate through /mcp. Cursor, VS Code, Cline, Windsurf, and other MCP clients can add it as a remote Streamable HTTP server.

Claude users can also install the supplied plugin through the Claude plugin marketplace. The plugin provides an AI visibility audit and a fix-planning skill, and its setup then asks the user to select the Proofsource MCP server and sign in.

New accounts using a work email receive two free scans. Personal email addresses can create an account but do not receive scans.

Tools and capabilities

Read tools cover:

• Workspace brands and tracked topics
• Visibility by engine or buyer question
• Competitors and comparative share of voice
• Cited websites and the brand site's citation position
• Daily trends, answer changes, and new category pages
• Open fix opportunities and fix verification
• Initial scan progress and finished report links
• An unsigned example report

Write tools cover adding a brand, creating a buyer question, and marking a fix as shipped. These actions apply to the user's workspace and are restricted by the account's plan and role. Repeating an identical brand addition, prompt, or shipped fix does not create a duplicate record.

Limitations and notes

The initial site-reading step can take one to two minutes. A scan must finish before its visibility results are available, and report links are generated from the newest completed report. Added questions run from the next engine run rather than changing past measurements.

Fix verification depends on subsequent samples and may remain awaiting samples or fall within normal answer churn. The service measures the engines and questions configured in the Proofsource plan; adding questions counts toward the plan's limit.