Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@doc-index search my documents for 'how to configure authentication'"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A local-first semantic search server for your documents. Index PDFs, Word docs, PowerPoints, Excel files, and text/markdown, then search them using natural language via the Model Context Protocol (MCP).
Semantic search - Find relevant content using natural language queries
Boundary-aware chunking - Respects document structure (chapters, sections, headers)
Table extraction - Extract tables from documents as CSV
Fully local - No external APIs, no cloud services, no Docker containers, no PyTorch
Lightweight - ONNX-based embeddings (~50MB vs ~2GB for PyTorch)
Related MCP server: KnowledgeMCP
Requires uv. If you don't have uv, see Alternative Installation below.
Add to .mcp.json in your project root (for Claude Code) or your Claude Desktop config:
{
"mcpServers": {
"doc-index": {
"command": "uvx",
"args": ["doc-index-mcp"]
}
}
}The skill teaches Claude how to use the search tools effectively (token budgets, boundary expansion, etc.):
uvx --from doc-index-mcp doc-index-install-skillThat's it — start asking Claude to index and search your documents.
Format | Extensions | Notes |
Text |
| Plain text |
Markdown |
| Preserves headers for boundaries |
| Text extraction with page markers | |
Word |
| Paragraphs, headings, tables |
PowerPoint |
| Slides, notes, tables |
Excel |
| Sheets as tables |
Component | Traditional RAG | This Server |
Embeddings | OpenAI API / hosted model | Local ONNX model (fastembed) |
Vector DB | Pinecone / Weaviate / Qdrant | Local file (usearch) |
Storage | Cloud / managed DB | Local |
Dependencies | PyTorch (~2GB) | ONNX Runtime (~50MB) |
doc_indexIndex a document for semantic search.
{
"file_path": "docs/manual.pdf",
"source_name": "manual"
}doc_searchSearch indexed documents using natural language.
{
"query": "how to configure authentication",
"top_k": 5,
"expand_to_boundary": "section",
"max_return_tokens": 4096
}Parameters:
query - Search query
sources - Filter to specific sources (optional)
top_k - Number of results (default: 5)
expand_to_boundary - Expand results to full "chapter", "section", "subsection", or "page"
max_return_tokens - Token budget for results (default: 4096)
include_siblings - Include sibling sections when expanding
doc_listList all indexed sources.
doc_chunkRetrieve a specific chunk by ID with optional neighbors.
{
"chunk_id": "manual:42",
"neighbors": 2
}doc_tocGet the table of contents (chapters, sections, subsections) for an indexed document. Use this to understand document structure before retrieving specific content.
{
"source_name": "manual",
"max_depth": 3
}doc_get_contentRetrieve document content by structural location. Provide exactly one locator: boundary_id, chapter, section, or pages.
{
"source_name": "manual",
"chapter": "3",
"max_return_tokens": 8192
}read_documentRead a document without indexing. Returns formatted text.
{
"file_path": "report.pdf",
"max_chars": 100000
}list_tablesList all tables in a document.
{
"file_path": "data.xlsx"
}extract_tableExtract a specific table as CSV.
{
"file_path": "data.xlsx",
"table_index": 0,
"max_rows": 100
}Variable | Description | Default |
| Base directory for resolving file paths | Current working directory |
| Directory for storing vector indices |
|
pip install doc-index-mcpThen in your .mcp.json:
{
"mcpServers": {
"doc-index": {
"command": "doc-index-mcp"
}
}
}Clone the repo and install dependencies:
git clone https://github.com/mike-anderson/doc-index-mcp.git
cd doc-index-mcp
pip install -e .Then point your .mcp.json at the server entrypoint:
{
"mcpServers": {
"doc-index": {
"command": "python",
"args": ["/path/to/doc-index-mcp/src/server.py"]
}
}
}Everything runs locally - no external APIs, databases, or embedding servers required.
flowchart TB
subgraph Client["MCP Client (Claude Desktop, etc.)"]
LLM[LLM]
end
subgraph MCP["Doc Index MCP Server"]
Server[server.py]
subgraph Services["Local Services"]
Loader[Document Loader<br/>PDF, DOCX, PPTX, XLSX]
Chunker[Boundary-Aware<br/>Chunker]
Embedder[Embedder<br/>ONNX Runtime]
VectorStore[Vector Store<br/>usearch]
end
end
subgraph Storage["Local Filesystem"]
Docs[(Source<br/>Documents)]
Index[(".docindex/<br/>├── manifest.json<br/>└── vectors/<br/> ├── index.usearch<br/> ├── chunks.jsonl<br/> └── boundaries.json")]
end
subgraph Models["Embedded Model (downloaded once)"]
ONNX[BAAI/bge-small-en-v1.5<br/>ONNX format ~50MB]
end
LLM <-->|MCP Protocol| Server
Server --> Loader
Server --> Chunker
Server --> Embedder
Server --> VectorStore
Loader -->|read| Docs
VectorStore <-->|read/write| Index
Embedder -->|load once| ONNX
style Client fill:#e1f5fe
style Storage fill:#fff3e0
style Models fill:#f3e5f5
style MCP fill:#e8f5e9flowchart LR
subgraph Index["Indexing"]
direction TB
A[Document] --> B[Load & Extract Text]
B --> C[Detect Boundaries]
C --> D[Chunk ~256 tokens]
D --> E[Generate Embeddings]
E --> F[Save to Disk]
end
subgraph Search["Searching"]
direction TB
G[Query] --> H[Embed Query]
H --> I[Vector Similarity Search]
I --> J[Expand to Boundaries]
J --> K[Return Results]
end
Index -.->|stored in .docindex/| SearchMIT
This server cannot be installed
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
Local-first RAG engine with MCP server for AI agent integration.
A comprehensive Model Context Protocol (MCP) server that enables AI assistants to interact with yo…
Agentic search over your Dewey document collections from any MCP-compatible client.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/mike-anderson/doc-index-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server