Skip to content

Release v0.5.0

Latest
Latest

Choose a tag to compare

View all tags
@thepushkaraj thepushkaraj released this 26 Jun 18:26
v0.5.0
a7d607d

contextmcp v0.5.0

This is a minor release. It adds three new embedding providers, two new documentation sources/parsers, and two new diagnostic commands. All changes are backward compatible — existing OpenAI/Gemini + GitHub configs keep working unchanged.

Features

Embedding providers

  • Cohere and Voyage AI are now supported embedding providers, called via their REST APIs with no new SDK dependencies.
  • Ollama is supported for fully-offline, no-API-key, zero-cost local embeddings against a local Ollama server. Set embeddings.provider: ollama (optional embeddings.ollama.baseUrl, default http://localhost:11434).

Sources & parsers

  • GitLab source — index docs from gitlab.com or a self-hosted GitLab instance via type: gitlab (set GITLAB_TOKEN for private projects, gitlabHost for self-hosted).
  • HTML parser — index raw .html/.htm docs (parser: html); strips boilerplate and converts to Markdown before chunking.

Commands

  • validate — checks provider / model / dimension / env-var consistency before a reindex, so misconfigurations fail fast. Now validates model dimensions for the hosted providers (OpenAI, Gemini, Cohere, Voyage), catching mismatches like cohere + embed-v4.0 + dimensions: 3072.
  • doctor — environment readiness checks (config parse, provider key presence, keyless handling for Ollama).

Fixes & Hardening

  • Network resilience: hardened error handling and retries across the fetchers and embedding calls — transient 429/5xx responses are retried with backoff.

Full changelog: v0.4.0...v0.5.0

Assets 2
Loading