v0.1.0 · MIT License
Enables an AI assistant to deterministically score hands of a Persian trick-taking card game from conversational bids and results, returning updated totals, validation errors, and plain-English explanations.
v0.1.1 · MIT License
Enables mining Amazon product reviews for one ASIN per request — fetching paginated reviews, surfacing complaint signals from critical reviews, summarizing sentiment and rating trends over time, ranking complaint themes, and producing a combined per-ASIN report across 14 marketplaces.
v1.0.0 · MIT License
Enables two AI agents to share a persistent task mailbox over remote MCP, so one can leave work, the other can claim and complete it, and results are written back with optional webhook wake-ups.
v0.2.3 · MIT License
Enables large language models to drive ZWCAD 2D and ZWCAD Mechanical through natural language, exposing 39 tools for drawing, dimensioning, layer/block/style management, and mechanical frames, title blocks, and BOM tables.
— · MIT License
Lets Claude Code sign in through Microsoft Entra ID and call infrastructure tools, while privileged actions stay blocked until a human approves that exact request on a web page. Grants are bound to the user, tool, and arguments, expire automatically, and can be revoked, so a valid token alone never grants standing access to production changes.
— · MIT License
Enables Claude to offload bulk reading, first-draft writing, and mechanical edits to a local or OpenAI-compatible model, with code-enforced verification, staged patches, project checks, and token budgets to save tokens and cost.
— · MIT License
Turns a HAR capture of an undocumented web system into an MCP server whose tools automatically handle hidden CSRF tokens, ETags, and rotating nonces, with write actions gated behind confirmation and exercised only against a sandbox. Verified tools are shipped alongside an honest report of the actions that could not be automated and still need a human.
— · Apache 2.0
Enables exhaustive, unit-by-unit analysis of open-source repositories through MCP, cataloging every file and symbol, resolving dynamically built routes and URL paths, and mapping call and value flows via LSP and Joern. Every conclusion carries file:line evidence tied to a pinned commit, with unresolved items tracked in an unverified list and assigned verification grades L0–L2.