Releases: jorgell23-sys/mdcx
Release list
mdcx 1.0.4
Renames the MCP server so it can be found.
The registry indexes by name, not by description. Published as mdcx, the server appeared only when searching for its exact name. It is now io.github.jorgell23-sys/markdown-document-search, which answers searches for markdown, document and search.
No functional change to the package.
mdcx 1.0.3
Fixes the MCP server serving only one call per restart.
The connection was cached and consumed from the thread pool the MCP SDK uses for synchronous handlers, so every call after the first failed. Handlers are now async, the connection allows cross-thread use, and every access is serialised: allowing cross-thread use without a lock returns wrong rows without raising.
Also fixes the document tool, which queried a column name the schema did not have and had never worked. Packages created by earlier versions remain readable.
49 tests.
mdcx 1.0.2
Align the three registries.
The Zenodo record title came from CITATION.cff, which had been left in Spanish while everything else was translated. It is now in English, and the concept DOI is recorded in the citation file and shown in the README.
mdcx 1.0.1
Refuse to write an empty package.
pack wrote a package with no documents without complaint, and the mistake surfaced only when querying it. The usual cause is pointing at the folder of source documents rather than at the converted Markdown. It now refuses and names the likely cause.
45 tests.
mdcx 1.0.0
First public release.
Converts document collections to verified Markdown, packages them into a single encrypted .mdcx file with search index and provenance, and exposes them to agents through the Model Context Protocol.
- 99.948% coverage over the collection used during development
- Answering one query costs 435 model tokens instead of 2,265,488
- Ed25519 signing so the issuer can be proven, not merely declared
- 43 tests over hostile inputs
- Querying installs ~62 MB; the conversion stack is a separate extra