A local-first semantic memory system that enables document ingestion, semantic querying, and knowledge graph traversal via MCP.
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.
Disclosed vulnerabilities in this server's declared npm dependencies (via OSV). Whether each is reachable depends on the installed versions.
ws: Memory exhaustion DoS from tiny fragments and data chunks
JS-YAML: Quadratic-complexity DoS in merge key handling via repeated aliases
uuid: Missing buffer bounds check in v3/v5/v6 when buf is provided
ws: Uninitialized memory disclosure
process.env. You'll be asked to provide them before it can run.CHUNK_OVERLAP— 64 Overlap between adjacent chunksCHUNK_SIZE— 512 Target chunk size in tokensCHUNK_SUMMARY_PROMPT_FILE— Per-chunk summary promptDATABASE_URL— "": "postgres://vkb:vkb@localhost:5433/vkb",EMBED_DIM— 768 Embedding dimension (must match model output)EMBED_MODEL— "": "nomic-embed-text",INGEST_MAX_RETRIES— 3 Max times a failed job is re-queuedIVFFLAT_LISTS— 100 Number of ivfflat lists (tune alongside chunk count)IVFFLAT_THRESHOLD— 1000 Chunk count above which an ivfflat index is created/maintainedJOB_TTL_DAYS— 7 Days before completed/failed jobs are expiredLLM_EXTRACT_CANDIDATES— 20 Candidate chunks the LLM considers per extraction passLLM_MODEL— "": "gemma4:e4b",LLM_RELATION_EXTRACTION— true Use LLM to extract relations (set false to use heuristics only)LOG_LEVEL— info Log verbosity: debug, info, warn, error. Pass --debug at startup as a shorthand for LOG_LEVEL=debug — also enables full request/response logging on all HTTP endpoints and MCP tool call tracingMCP_PORT— vkb uses stdio transport when is 0. The MCP server is also accessible over HTTP Streamable (MCP 2025-03-26 spec) when MCP_PORT > 0 (default 3333), with session management for multi-client use.MCP_SECRET— Bearer token for the HTTP MCP endpoint on MCP_PORT (leave unset to disable)MCP_STDIOOBS_PORT— The observability server runs on (default 4242) and exposes both the SPA and a REST API.OBS_SECRET— Set =your-secret in .env. When set, all API requests must include:OLLAMA_BASE_URL— "": "http://localhost:11434",RAWSTORE_ADAPTER— filesystem Raw content storage: filesystem or s3RAWSTORE_PATH— ./rawstore Root path for filesystem rawstoreRAWSTORE_S3_BUCKET— S3 bucket name (when RAWSTORE_ADAPTER=s3)RAWSTORE_S3_ENDPOINT— S3-compatible endpoint URL (optional override)RELATION_CONFIDENCE_STEP— 0.05 Increment applied on relation confirmationRELATION_EXTRACT_PROMPT_FILE— Relation extraction promptRELATION_PRUNE_THRESHOLD— 0.6 Confidence below which relations are pruned on retuneRELATION_THRESHOLD— 0.75 Minimum cosine similarity to create a relationRELATION_TOP_K— 10 Nearest neighbours considered per chunkRELATION_TTL_DAYS— 30 Days before unconfirmed relations are prunedRETUNE_INTERVAL_HOURS— 6 How often the retune worker runs automatically (0 = disabled)RETUNE_SUMMARISE— false Regenerate entity summaries during retuneSECTION_MAX_SIZE— 8 Max chunks per sectionSECTION_SPLIT_THRESHOLD— 0.65 Cosine similarity drop that triggers a section boundarySECTION_STRATEGY— similarity_valley Sectioning strategy: similarity_valley or positionalSECTION_SUMMARY_PROMPT_FILE— Section summary promptSECTION_WINDOW_SIZE— 5 Sliding window size for valley detectionSUMMARY_CONCURRENCY— 4 Parallel LLM calls during the summarising stageSUMMARY_MAX_INPUT_CHARS— 12000 Max characters fed to the entity-level summary prompt (~3 k tokens)SUMMARY_PROMPT_FILE— Entity summary promptTLS_CERT— Path to TLS certificate file (enables HTTPS/WSS on all servers)TLS_KEY— Path to TLS private key fileWORKER_CONCURRENCY— 2 Number of ingest worker processes[](https://m8ven.ai/mcp/inkiebeard-galactic-vkb-c8gpsp)