peopleworks/SignsofAI

Explainable AI-writing detection and originality checking for academic integrity, in English and Spanish. `analyze_ai_writing` returns a 0–100 score plus every tell behind it — overused vocabulary, rhetorical crutches, syntactic patterns, and sentence-rhythm burstiness — each with the matched text and a fix, instead of a black-box percentage; it also publishes its own error rate rather than asking to be trusted, flagging 2 of 296 texts written before 2022. `check_originality` returns the shared passages between documents as evidence; `check_citations` finds where a document contradicts its own bibliography; `compare_to_baseline` sits a piece against the same writer's earlier work; `inspect_characters` surfaces invisible and lookalike characters; `write_report` exports the whole analysis as a document a person can take to a committee. Every tool but two runs entirely on the machine and the text never leaves it; the two server-backed ones (perplexity, cross-language paraphrase) disclose that they send text. stdio transport, .NET 10, MIT. Install: `dnx SignsOfAI.Mcp --yes`, or `dotnet tool install --global SignsOfAI.Mcp`. Browser demo: https://peopleworks.github.io/SignsofAI/

Project

Browse this MCP server in the TensorBlock MCP Index.