Snapshot-based editing for AI agents: send only the NEW content via stdio MCP. Immutable sha256 snapshots, 3-way merge, conflict detection, ambiguity-safe zero-write applies, cross-process locks.
A small, extensible harness for building test MCP servers that reproduce common MCP attack patterns, so you can measure whether your detector catches them. Built on the official MCP Python SDK (mcp 2.x).
MCP server with a context ledger: cryptographically signed receipts of exactly which file versions the model was shown. Honest benchmarks: no real-world token savings (-1.4% measured); ~72% applies only to a synthetic re-read loop.
Ask Cooter ingests a PDF, uses a vision model to pull clean text, tables, and figure descriptions from every page, so it works on scanned documents with no selectable text, and embeds it into pgvector. Ask questions and get answers with citations back to the exact source page, via a CLI, a local streaming chat UI, or an MCP server.
Enterprise multi-agent financial analytics platform leveraging Anthropic's Model Context Protocol (MCP) and Claude 3.5 Sonnet for automated text-to-SQL execution, market data integration, and self-healing validation loops.