Knowhere MCP
The MCP server lets agents explore prepared documents, sections and related assets while retaining source references.
- 3,911 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The server stores uploaded documents and sends content to selected models for processing and retrieval.
Install
- Works in
Manual install
uv sync --all-packagesFirst Quick Start command in the downloaded repository. API/worker configuration, external services and migrations follow.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server stores uploaded documents and sends content to selected models for processing and retrieval.
This is third-party code. Review the repository files before installing.
What it does
Knowhere MCP connects an agent to a prepared document collection. The agent explores outlines, finds sections, reads passages and follows related assets. Evidence retains links to its document, page and source image where available. Knowhere processes uploaded files before retrieval. The newer API uses a vision model for PDFs and PowerPoint, while text documents follow another path. Both feed the same retrieval tools.
Who it is for. Analysts of long reports and technical archives who need answers traceable to a specific source page.
Good fit when
- A long PDF contains a table in a known section.
- Several documents need cross-referenced passages on one topic.
- A conclusion needs the original page image for verification.
- Repeated sessions use the same prepared document collection.
Not a fit when
- Documents must not reach the selected model provider.
- Audio or video must be handled as a supported format.
- A one-off lookup cannot justify running the service and its storage dependencies.
Example request
Find the cost table in chapter three of the report in Knowhere and show the source page.MCP
- Transport
- http
- Authentication
- API key
Limitations
Self-hosting requires Python 3.11 or later, uv and Docker with Compose; prepare that environment first. A database, Redis and S3-compatible storage are also needed, beyond Python dependencies. Page processing and retrieval use selected models that receive document content, so check provider terms for confidential files. Product telemetry is enabled by default and can be disabled with `TELEMETRY_ENABLED=false` before use.
How to disable. Remove the client MCP connection and revoke its API key. Back up documents, database records and object storage separately before stopping a self-hosted service.
FAQ
Is MinerU required for PDFs?
No. Its optional page-text extraction stage is off by default.
Where are local API docs?
Open `http://localhost:5005/docs` after starting the API.
Does this repository also start the dashboard?
No. Knowhere Dashboard is run separately and connects to the API.
Can a previously read section be excluded?
Yes. The agent can exclude documents and sections from a subsequent query.
Why process files before querying?
The server searches prepared documents. Wait for processing to finish before checking answers.
Discussion
No questions or reviews yet.
Related
MCP serversThe MCP server lets agents read and search local Molio vaults, browse file trees and ask questions about their contents.
The MCP server connects an agent to a learning canvas where a question about a passage opens a linked document.
The MCP server answers questions from a personal document collection and returns linked passages with sources.
MCP PDF for large documents: the agent searches by meaning and keyword across one file or a folder and reads only the pages, tables and scans it needs
PageIndex MCP: long PDFs for an agent
PageIndex MCP
An MCP server that lets an agent read long PDFs through a document tree: retrieval by reasoning instead of a vector database
A local document base for your agent: hybrid semantic and keyword search plus a web UI for browsing and uploads
README in short
Text and page processing share a hierarchy and evidence-reference schema. Entities and keywords link related documents.