Skip to content

Releases

Readest MCP v0.1.1

Choose a tag to compare

@gbettineschi gbettineschi released this 27 Oct 21:41
92fbecd

Fixes for annotation filters and process-crash recovery.

  • Date filters compare actual instants across UTC offsets and fractional-second formats, including finer-than-millisecond lower bounds.
  • A persistent SQLite guard prevents competing dead-owner recoverers from removing a newly acquired lock. Existing PID ownership remains for cooperative prior-release clients.
  • Expanded filter, annotation/vector lifecycle, interrupted cache transaction and cross-process tests. CI now runs the real local multilingual model and Tesseract OCR on macOS, Linux and Windows.

Validation: 139 tests passed locally on Node 24.15.0 and 24.22.1 with real model/OCR/account checks. CI omits the private-account test. A separate real Readest highlight/comment was created and edited, retrieved through MCP, matched to its source context and checked against date bounds. See the compatibility evidence for test boundaries.

Upgrade: restart Readest MCP in every connected client so existing processes load the fixes. Private generated SQLite guard files remain beside cache/session locks; do not remove them while clients are running. The public six-tool read-only interface is unchanged.

Download the compiled readest-mcp-0.1.1.tgz and SHA256SUMS. With Node 24.15+ in the 24 LTS line:

npm install ./readest-mcp-0.1.1.tgz
npx readest-mcp --help

Model files and Tesseract remain separate optional installs. The package contains no personal books, account credentials, models or native OCR executables.

Readest MCP v0.1.0

Choose a tag to compare

@gbettineschi gbettineschi released this 27 Oct 20:39
c6ffce5

First release of Readest MCP: a small, read-only server built with the official TypeScript MCP SDK.

  • The same six tools for Readest Cloud and saved local libraries: browse books and annotations, search, and retrieve cited EPUB/PDF passages.
  • Keyword search and optional local multilingual semantic search using continuous section context and retrievable matching windows.
  • Native PDF text, conservative blank-page detection, and optional bounded local Tesseract OCR with explicit provenance.
  • Source/account-separated caches, recoverable atomic cache/session locks, stale-reference handling, and explicit availability and coverage reporting.

Download readest-mcp-0.1.0.tgz and SHA256SUMS. With Node 24.15+ in the 24 LTS line, install the package in your chosen directory:

npm install ./readest-mcp-0.1.0.tgz
npx readest-mcp --help

See the README for configuration and client setup. Model files and Tesseract are separate optional installs; book text is not sent to an embedding/OCR service. The package contains no personal books, credentials, models or native OCR binaries. No npm registry publication is needed for this release.

Validation: 107 tests passed locally with the real model, OCR and live account enabled. CI covers macOS, Linux and Windows. All six tools and exact-reference round trips were exercised on 26 private books; six paired Italian/English queries found substantive expected sources in the top five. These checks do not establish exhaustive semantic recall or perfect OCR accuracy.

Built for personal use and shared as is; development follows personal needs and available time. Thoughtful fixes and contributions are welcome.