Citation Check

What Citation Check MCP server does

Citation Check evaluates citations against two separate questions: whether the cited URL exists, and whether the source identifies the same entity named by the claim. Keeping these checks separate helps expose citations that are reachable but point to the wrong subject. For example, a URL may return successfully while discussing a different gene from the one named in the claim.

The service is intended for agents or workflows that need to inspect cited web sources. It does not reduce the result to URL availability alone. A reachable page can still fail the entity comparison, so both dimensions should be read independently.

How it works

The Citation Check MCP server receives a URL and the claim associated with that citation through the check_citation tool. It fetches the URL, reports whether the URL exists, and checks whether the source names the same entity as the claim. These are distinct outputs rather than one combined test.

Use verdict_meanings to read the precise definitions for every status and relevance value. This reference also identifies what the checker cannot determine. Consult it before interpreting automated results, especially when a workflow needs to distinguish an unavailable source from a source that is available but does not support the named entity.

The available material does not specify a transport mode, installation command, client configuration, or local runtime requirements. It identifies the service at https://exmachina-ai.fly.dev/verify and describes its MCP tools, but does not provide further deployment instructions.

Tools and capabilities

• check_citation: Fetches a cited URL and reports URL existence plus whether the source names the entity named in the attached claim.
• verdict_meanings: Explains the semantics of all returned status and relevance values, including the checker’s stated boundaries.

The Citation Check MCP server requires no account according to the supplied description. It also states that nothing about the user is stored. Those properties make it suitable for checking citations without adding account setup to the described workflow, but they do not imply that page content is retained or that every claim can be fully validated.

Limitations and notes

A successful URL response is not evidence that the citation refers to the correct entity. The checker exists specifically to distinguish those cases. Read the relevance interpretation separately from the existence result, and consult verdict_meanings before assigning your own meaning to a status.

The supplied material does not describe authentication options, request parameters beyond a URL and its attached claim, response schemas, rate limits, supported page types, or installation steps. It also does not state that the tool proves a claim’s broader factual accuracy. The documented comparison concerns whether the source names the same entity as the claim; use the documented verdict semantics to understand cases the checker cannot determine.

For integration decisions, treat the endpoint and two named tools as the confirmed interface. Additional behavior should not be assumed without more documentation.