io.github.sfrangulov/minirag-mcp
Local-first RAG MCP server: hybrid search over a folder of your own documents
The current version, v0.6.1, was published to the official MCP registry on 2026-09-07. It is distributed as minirag-mcp on PyPI, filed under Filesystem by this index, and declares 13 environment variables. Removals and unreachable sources are measured across every server this index tracks, contract drift across the servers that answer in consecutive snapshots; the index's current counts put this record in context.
Using io.github.sfrangulov/minirag-mcp in Claude, Cursor, Gemini CLI, Cline, or Zed?
MCP tool contracts can change remotely with no version bump. The mcpindex gate pins each contract and HOLDs the call when it drifts-before your agent acts. Zero credentials. This is not the package install for this server itself (use Install this server for that).
Rewrites your MCP host config so each server launches behind the gate. Inspect first: curl -fsSL https://mcpindex.ai/install.sh | less
uv tool install mcpindex-gate && mcpindex-config-wireSemantic screen found no manipulation pattern in the description. Conformance probe not yet run.
mcpindex.integrity.descriptionpassINFOevidence“No malicious instructions found”via static_description
- - Semantic screen only - the deterministic conformance probe has not run on this server
- - Confidence is reported but not yet calibrated (v1)
- - Screen reads the tool description, not the live behavior
- - advisory
- - registry description only no input schema
- - screen model 8b
- - Still current: the description we screened is unchanged, re-checked against the registry
- - The registry listing (version, packages, links) changed after this screen ran - the description we assessed did not
Semantic screen: an LLM judge reads the tool description for hidden instructions (status PARTIAL). A pass means the description is not lying, not that the tool is safe: a high-capability tool with an honest description still warrants caution. The deterministic conformance probe has not been run on this server yet, so the screen here is semantic-only. Posture: advisory. Confidences are reported but not yet calibrated (calibrated=false at v1). Full verdict history is not shown on this page.
Own this server? Screen its description →
That verdict was true at screening time (snapshot 2026-09-23).
Contracts can change after screening, with no version bump. The gate pins io.github.sfrangulov/minirag-mcp’s tool contracts on first sight and holds any silent change before your agent acts - the check that keeps being true on Tuesday.
See your first HOLD in 2 minutes →
Related: how to trust an MCP server · screen before install · silent contract drift
People vet a server before wiring it in. One line on your project site or docs answers that with an independent record: screening verdict, registry provenance, contract drift history. It stays current as new screens and drift events land.
[Independent trust record for io.github.sfrangulov/minirag-mcp](https://mcpindex.ai/server/io-github-sfrangulov-minirag-mcp) - screening verdict, registry provenance, and contract drift monitoring.<a href="https://mcpindex.ai/server/io-github-sfrangulov-minirag-mcp">Independent trust record for io.github.sfrangulov/minirag-mcp</a> - screening verdict, registry provenance, and contract drift monitoring.A live verdict badge for your README or listing. It reflects the current screen, links back here, and updates when the verdict does.
[](https://mcpindex.ai/server/io-github-sfrangulov-minirag-mcp)<a href="https://mcpindex.ai/server/io-github-sfrangulov-minirag-mcp"><img src="https://mcpindex.ai/api/v1/badge/io-github-sfrangulov-minirag-mcp" alt="mcpindex verdict" height="20" /></a>BASE_DIROne document root; also the security boundary for file access. Defaults to the process working directory. Ignored when BASE_DIRS is set.
BASE_DIRSJSON array of document roots, e.g. ["/docs/a", "/docs/b"]. Takes precedence over BASE_DIR. An invalid value is a hard configuration error.
DB_PATHLanceDB index directory. Defaults to <first root>/.minirag/lancedb, so each corpus gets its own index.
CACHE_DIREmbedding model cache. Defaults to the platform user cache dir, so the ~220 MB model is downloaded once and shared.
MODEL_NAMEfastembed model id. Changing it makes existing vectors incompatible with new queries; pair with a new DB_PATH or a full re-ingest.
MAX_FILE_SIZEPer-file size limit in bytes, enforced before parsing.
CHUNK_TOKEN_BUDGETRetrieval-unit size in the embedding model's own tokens. Range 16-128; the ceiling is the model's trained sequence length.
RAG_HYBRID_WEIGHTKeyword weight in the weighted RRF fusion, range 0.0-1.0. 0 is vector-only; higher values raise the BM25 contribution.
RAG_GROUPINGResult grouping filter. 'similar' keeps only the closest group; 'related' also keeps the next one. Unset means no grouping filter.
RAG_MAX_DISTANCEDrop results whose vector distance exceeds this value. Lower is stricter. Unset means no distance filter.
RAG_MAX_FILESKeep chunks from at most this many best-scoring files. Unset means no per-file filter.
RAG_INSTRUCTIONS_APPENDExtra paragraph appended to the instructions the server hands the client at connect time, for corpus-specific guidance.
ALLOW_PRIVATE_URLSLet ingest_url fetch hosts resolving to loopback, link-local, private, reserved or unspecified addresses. Off by default.
Track SaaS directory submissions for your products via API key–gated Streamable HTTP MCP.
Issue tracker and durable project memory for your coding agent, scoped to the folder you open.
Filesystem-direct Obsidian MCP server — search and edit your vault with low context-tax.