MCP server

Teams increasingly fix sites together with AI assistants such as Claude Code, Cursor and Windsurf. The Pandus MCP server gives the assistant direct access to your Pandus data: it sees what to fix, where the problem is on the site and how to fix it from the rule card, and after the change it starts a re-check by itself.

What the assistant can read

• the sites of your team;
• what to fix first for a site;
• a problem card: the rule, the WCAG criterion, pages, selectors, an HTML example and how to fix it on the platform;
• live barriers of the SDK for a page or a template;
• a short analytics summary: visits, conversion, barriers on the funnel steps;
• the history of checks and the comparison “before and after”.

What it can do, only with the right in the token

• start a check of a site or a page;
• mark a problem as fixed and start the confirmation;
• create an event without code.

Resources and prompts

• Resources: the rule cards in seven languages and the scoring method.
• Prompts: “fix the top 3 problems on this page” and “prepare a report for the client”.

Tokens and rights

The server works on top of the public API /api/v1 with the same tokens. Create a token in API and webhooks → MCP and give it only the rights it needs:

A token also lists the sites of the team it may see.

Connecting

The server runs remotely over streamable HTTP at mcp.pandus.online, and as a local package over stdio. API and webhooks → MCP in the dashboard gives ready configurations for Claude Desktop, Claude Code and Cursor to copy.

Limits and safety

• Only the data of your own team; team isolation is covered by tests.
• No personal data of visitors: only SDK aggregates, never session recordings.
• The same request limits as the API.
• Every call is written to the team's activity log.
• No action without the right in the token.