Vaelor is a self-hosted MCP server that parses 16 languages into a call graph and a knowledge graph, so a coding agent sees callers, blast radius and prior review findings before it edits. Apache-2.0. Site: vaelor.ai. Source: GitHub.
What it does
- MCP tools for impact analysis, call tracing, semantic search, dead-code detection and PR review (35+ tools, per vaelor.ai on 2026-11-19).
- A persistent graph on PostgreSQL with Apache AGE. PageRank, community and surprise scores feed review and impact answers.
- A review memory loop:
review_prstores one verdict per changed symbol from the real GitHub review outcome, andunderstandreturns it on the next call. debug_investigatejoins a Prometheus alert, failed Jaeger traces and code symbols into rankedfile:functionsuspects.- Hybrid retrieval: BM25F, 768-dimension pgvector embeddings, one-hop graph expansion and a reranker.
Status
First commit 2026-02-27, in active development. The latest release on 2026-11-19 is v1.76.0, after 223 version tags: that counts releases, not adoption.