A self-hosted semantic code retrieval MCP server that enables natural language queries to find relevant code snippets using Tree-sitter AST parsing, embedding, and hybrid search with reranking.
Claim it to get a verified publisher badge, a free copy of our full audit findings, and direct contact for any high-priority issues we find.
Install from
M8ven verifies MCPs across every public registry — install directly from whichever one you prefer.
process.env. You'll be asked to provide them before it can run.SCM_EVAL_TARGETSCM_RERANK_BACKEND— auto rerank 后端：auto（有 VOYAGE key 用 Voyage，其次 Cohere）/ voyage / cohere / offSCM_EMBED_BACKEND— local # 本地 embedding（默认 Qwen/Qwen3-Embedding-0.6B）SCM_EMBED_CHAR_BUDGET— 200000 单批最大字符数（未绑卡时可调小）SCM_EMBED_DTYPE— float 向量精度：float（最高精度）/ int8（省 71% 存储）SCM_EMBED_MAX_RETRIES— 4 API 限流/错误时自动重试次数VOYAGE_API_KEY— 「效果优先」技术栈：Tree-sitter AST 切分 + voyage-code-3 embedding（int8 量化）+ sqlite-vec 向量库 + BM25 全文 + RRF 融合 + Voyage rerank-2.5 重排 + Call Graph 图扩展。一个 跑通 embedding + rerank 全链路。SCM_EMBED_MIN_INTERVAL— 0 请求最小间隔秒数（未绑卡 3 RPM 限制时设 21）SCM_LOCAL_EMBED_MODELSCM_LOCAL_EMBED_DEVICESCM_LOCAL_EMBED_MAX_SEQSCM_LOCAL_EMBED_BATCHSCM_LLM_BASE_URLSCM_LLM_API_KEYSCM_LLM_MODELSCM_LLM_TIMEOUTSCM_QUERY_EXPANSION— 自配 LLM 端点（可选，默认关） 仅查询文本（不含代码） =on 时SCM_EMBED_CONCURRENCY— voyage 3 / local 1 索引时 embedding 并发请求数（实测 3 路≈提速 3 倍）SCM_RERANK_MIN_INTERVAL— 节流；429 自动退避重试兜底。SCM_RERANK_MODEL— 按后端 voyage → rerank-2.5，cohere → rerank-v3.5COHERE_API_KEY— 否 可选 rerank 后备后端；默认用 Voyage rerank-2.5，无需此 keySCM_LOG_FILESCM_DB_DIR— ~/.semantic-code-mcp 索引数据库存放目录SCM_LOG_DISABLESCM_LOG_MAX_BYTESSCM_LOG_STDERRSCM_EMBED_MODEL— voyage-code-3 Embedding 模型SCM_MAX_WORKSPACES— 8 最大同时管理的工作区数（LRU 淘汰）SCM_TOP_K— 50 向量召回候选数SCM_TOP_N— 10 最终返回结果数SCM_WATCH— 1 文件监控：1 启用（默认），0 禁用[](https://m8ven.ai/mcp/huawang1258-semantic-code-mcp-1meg72)