pyLLMSearch MCP
An MCP server gives agents search and answers over local documents using hybrid retrieval, reranking and section links.
- 662 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- MCP includes index updates and exposes source documents to the connected client.
- The default API listens on all network interfaces, so external restrictions are needed to keep data from other machines.
Install
- Works in
Manual install
FASTAPI_RAG_CONFIG="/path/to/config.yaml" FASTAPI_LLM_CONFIG="/path/to/llm.yaml" llmsearchapiStarts the API and MCP after installing pyLLMSearch; requires Python 3.10 or later. Substitute your YAML paths and restrict network access first.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- MCP includes index updates and exposes source documents to the connected client.
- The default API listens on all network interfaces, so external restrictions are needed to keep data from other machines.
This is third-party code. Review the repository files before installing.
What it does
pyLLMSearch MCP exposes a prepared document collection to an agent. It can return retrieved passages or a model answer with sources. YAML configuration defines the material, retrieval and generation setup. Supported formats include Markdown, PDF and DOCX, with optional processors extending the list. New documents can update an index without a mandatory full rebuild. The server also exposes an index-update operation, so a connected agent can change the local index.
Who it is for. For agent users who need configurable model-backed search over their own documents.
Good fit when
- Find answers in a local PDF and Markdown collection.
- Handle a question using different terminology from the document.
- Return source passages for separate answer verification.
- Update an existing index after adding documents.
Not a fit when
- You cannot prepare configuration and an index before querying.
- Documents must not reach the chosen cloud model provider.
- You need strictly read-only tools without index updates.
Example request
Connect MCP to the test document collection and return passages supporting this answer without updating the index.MCP
- Transport
- sse
- Authentication
- not required
Environment variables
FASTAPI_RAG_CONFIGrequired- Retrieval configuration path.
FASTAPI_LLM_CONFIGrequired- Model configuration path.
Limitations
Set the retrieval and model YAML paths before startup so the API can open the chosen collection. By default it listens on every interface at port 8000, with no established built-in authentication; deny access from untrusted machines. Local-model support does not make every setup offline: inspect the selected model endpoints before using private documents. The latest commit was in January 2026, so test search on a sample collection with your dependency versions.
How to disable. Stop `llmsearchapi` and remove the client connection. Indexes and source documents remain in their configured directories.
FAQ
Which MCP tools are exposed?
Four operations: passage retrieval, two answer variants and index updating.
How does the API start?
Run `llmsearchapi` after configuring the environment paths. The default port is 8000.
Which endpoint should the client use?
The documented local endpoint is `http://localhost:8000/mcp` over SSE.
Can results link to a document location?
Yes. Deep links can target a PDF page or Markdown heading.
How do I update an index manually?
Use `llmsearch index update -c /path/to/config.yaml` with your configuration path.
Discussion
No questions or reviews yet.
Related
MCP serversOfficial MCP server from the OpenSearch project exposes a cluster to agents: indexes, mappings, DSL search and shard state.
Elasticsearch MCP connects multiple search clusters, with an explicit target name and access scoped to the supplied credentials.
A local MCP server that turns a folder of documents into a knowledge base with hybrid search for Claude Code and Cursor
An MCP server for searching and downloading academic papers from arXiv, PubMed, bioRxiv, Semantic Scholar and other open sources
MCP server for the Flexible GraphRAG platform: ingest from 14 data sources, hybrid search and AI answers over a knowledge graph
A local knowledge base over PDFs, scans and Office files: hybrid search, access control and answers that cite the page
README in short
Retrieval can combine keywords and meaning, then rerank the returned passages. During a conversation it can reformulate a new question using earlier messages.