This repository was archived by the owner on Jun 12, 2026. It is now read-only.
Releases: er77/code-graph-rag-mcp
Releases · er77/code-graph-rag-mcp
Release list
v2.7.15
Release Notes (Consolidated) — @er77/code-graph-rag-mcp v2.7.9 → v2.7.15
This file consolidates recent release notes into a single document for easier tracking.
Install / Upgrade (latest)
- local artifact:
npm install -g ./er77-code-graph-rag-mcp-2.7.15.tgz(if present)
Node.js: >=24
v2.7.15 (2025-12-17)
- Maintenance release (version bump + rebuild).
v2.7.14 (2025-12-17)
- Indexing: include Markdown files (
.md,.mdx) in the parser/index pipeline. - New Markdown parsing: emits
document+headingentities andcontainsrelationships for heading structure.
v2.7.13 (2025-12-16)
Agent-quality upgrade for MCP tool usability and reliability:
- Tool guidance: rewritten tool descriptions with “Use when / Typical flow / Output” structure.
- Response standardization: unified JSON envelope via
toolOk/toolFail(includingagent_busyfailures). - Pagination: cursor +
pageSizesupport for high-cardinality tools (semantic_search,query,get_graph). - Schema/behavior alignment:
depthnow honored bylist_entity_relationshipsandanalyze_code_impact. - New grounding tools:
resolve_entityfor ranked disambiguation by name (+filePathHintboosting).get_entity_sourcefor snippet extraction with context + truncation safeguards.
- Query improvements: hybrid results include structural match annotations and paging metadata.
- Tests: added/updated integration + utility tests to prevent schema/behavior drift.
v2.7.12 (2025-12-15)
- Dependencies:
onnxruntime-nodeis now an optional peer dependency (avoids pulling deprecatedboolean@3.2.0).
v2.7.11 (2025-12-15)
- Database isolation: default
database.pathis./.code-graph-rag/vectors.db(per-repo storage by default). - Index hygiene:
.code-graph-rag/**is always excluded from indexing.
v2.7.10 (2025-12-15)
- sqlite-vec loading: load sqlite-vec via
sqlite-vec’sgetLoadablePath()first (global-install safe), with fallbacks.
v2.7.9 (2025-12-15)
Reliability release focused on stricter MCP clients and big repos:
- Stdio hardening: prevent stdout log pollution during MCP runs so strict clients can complete
initialize. - Logging: mirror server logs to
/tmp/code-graph-rag-mcp/mcp-server-YYYY-MM-DD.logfor early-start debugging. - Indexing defaults: exclude common build/tmp/vendor directories unless explicitly overridden.
- Batched indexing:
batch_index(resumable, progress-returning) to avoid strict client tool-call timeouts on big repos. - Incremental indexing:
incremental:truereindexes only changed files and safely replaces per-file graph rows. - Vector DB location: YAML supports
~and defaults were improved for predictable paths. - Graph query fix: RegExp name filtering supports exact
^...$and substring matches (unblocks relationship tools).