Skip to content

Releases: cioffiAI/mcp-agora

v0.4.1 - Bug fixes and improvements

Choose a tag to compare

@cioffiAI cioffiAI released this 10 Jul 07:35

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog,
and this project adheres to Semantic Versioning.

[0.4.1] - 2026-07-10

Fixed

  • agora_forget: Now properly clears both L1 and L2 caches (previously only L1 was cleared, leading to potential stale results after deletion).
  • agora_crossref (by entry_id): Now performs a direct ID-based retrieval from the ChromaDB collection instead of treating the opaque entry ID as a semantic query string. Cross references are now correctly derived from the actual stored document text.

Changed / Improved

  • Startup performance: Moved the heavy VectorStore (ChromaDB) import inside create_server() so that a bare import agora.server no longer triggers the expensive dependency load at module import time.
  • Improved test coverage for cache invalidation and cross-reference behavior using the real shipped tool implementations.
  • Added internal cache attachments (_agora_l1_cache, _agora_l2_cache) for reliable verification of L2 behavior in tests.
  • .gitignore updated to ignore mcps/ (tool schema captures, out of core scope).

Verification

  • All changes are covered by real tests that exercise the shipped code paths.
  • Explicit verification of L2 cache invalidation (pre-forget L2 hit → forget → post-forget miss with no stale data).
  • MCP handshake responsiveness test remains passing.

[0.4.0] - Previous release

Initial public release (see git history for details).

v0.4.0 — Phase 4 Robustness + Phase 5 Polish

Choose a tag to compare

@cioffiAI cioffiAI released this 06 May 13:59

What's New

Phase 4 — Robustness

  • Structured logging (file-based, configurable level via LOG_LEVEL env)
  • Backend health checks (healthy/unhealthy/dead with 3-failure threshold)
  • Configurable retry per backend (max_retries + retry_delay with exponential backoff)
  • Rate limiting per backend (rate_limit_rps token bucket)
  • Non-blocking startup + WarmingUpError grace

Phase 5 — Portfolio Polish

  • PDF tesina updated: v0.4.0, 61 tests, 10 files, Phase 4 completed
  • Config YAML fully commented (every field explained)
  • GitHub Actions CI (test + lint + format on push/PR, ubuntu + windows)
  • Tagged release v0.4.0

Changelog

439104f feat: Phase 5 portfolio polish
90765e8 chore: add *.log and chroma.sqlite3 to .gitignore
40e86f6 fix: non-blocking startup + WarmingUpError grace
c4aa01c feat: Phase 4 robustness
7f01873 docs: ARCHITECTURE.md rewrite, ruff linter, PDF tesina
dab75a2 feat: Phase 3 cross-agent memory
b03abbe fix: ReadOnlyBlockedError public, parallel broadcast
0c36648 feat: HTTP connector, read_only, broadcast, timeout
12e54ae docs: README + AGENTS.md update
5a64014 Phase 2: Routing + Backend Connectors
005590b Add AGENTS.md instructions
e209f40 Add MIT License
14dbdc5 Initial commit: Phase 1 complete

Tests

61 tests across 10 files, all passing. CI runs on Ubuntu + Windows.