Repository navigation
New: reusable document profiles. Save literal labels, required document anchors and field types once, then apply the profile to recurring forms/reports in one batch. Missing, invalid or ambiguous fields need review; partial values are withheld. Supports TXT/Markdown, DOCX and text-layer PDFs. No model training, API calls, automatic inference or telemetry. Eight MCP tools remain; the profile mode extends extract.
By AVRG3. This is opt-in: every new document is still parsed and checked. Prefer a direct read for one small file or an existing correct script. Smaller output is not proof of a lower bill; setup, agent turns and caching affect cost. Contract, examples and synthetic evidence: https://github.com/Warddamn/tiny-tools/blob/main/packages/context/docs/DOCUMENT_PROFILES.md
Choose the bundle for your OS: darwin = macOS, win32 = Windows, linux = Linux. Each contains x64/arm64 dependencies and requires Node.js 20+ or a compatible host-provided runtime. Open it in an MCPB-compatible client. The existing no-login npx installation remains in the README for other clients.
All eight tools are checked against known answers after extracting the actual bundle. These are unsigned bundles; the registry metadata includes their SHA-256 hashes. Dependencies are included, so bundles are larger than the npm-style download.
Tools have startup and validation costs. Task-level comparisons include both improvements and regressions; no universal savings are promised. Start with the worked examples: https://github.com/Warddamn/tiny-tools/blob/main/QUICKSTART.md