Uses Hugging Face as the source for its embedding and cross-encoder reranking models (defaults: Qwen/Qwen3-Embedding-0.6B and Alibaba-NLP/gte-reranker-modernbert-base), downloading them on first use and allowing any Sentence Transformers model to be configured via environment variables.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Local search over your documents, for AI agents. sift indexes folders of documents and notes (Markdown, Obsidian vaults, PDF, Office files, emails), embeds them on your GPU, and serves keyword, semantic and hybrid search to Claude Code, Cursor and other agents as an MCP (Model Context Protocol) server. Everything runs on your own machines: no cloud services, no API keys.
Four kinds of search: exact keywords (BM25), meaning (vectors), hybrid with cross-encoder reranking for RAG, and documents like a given one.
Your files as they are: Markdown with front matter, text, PDF (scans too, by OCR), Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, HTML, saved emails and JSON Lines, indexed incrementally. sift never writes to them.
Built for agents: filters on front-matter fields, results that cite their source lines, and compact results that save context. A note an agent just wrote is found by the next search.
Linked notes: in an Obsidian vault or any Markdown with [[wikilinks]], the links and backlinks of a note, the
notes not written yet, and the documents that mention a note without linking it.
A browser UI to search, browse and read the index yourself.
Measured, not guessed: the default models and ranking were chosen on seven retrieval benchmarks (benchmarks).
Runs on what you have: NVIDIA or Apple GPUs, the CPU, or another machine's GPU, on Linux, macOS and Windows. Tens of millions of passages fit on one machine, and keyword search answers over them in milliseconds.
Related MCP server: sage-mcp
Requires Python 3.11 or newer. With uv:
uv tool install --torch-backend auto sift-rag # installs the `sift` commandOr run it without installing: uvx --torch-backend auto sift-rag serve. With pip, in a virtual environment:
pip install sift-rag. On first use, sift downloads its two models from Hugging Face (about 2 GB).
GPU. sift uses an NVIDIA GPU (CUDA) if there is one, else an Apple GPU, else the CPU, which embeds slowly and
reranks in about 25 seconds per query on 8 threads (0.3 seconds on an RTX 4090). On Linux and macOS, PyTorch from PyPI
already uses the GPU. On Windows it doesn't: uv's --torch-backend auto picks the CUDA build that matches your
driver, and with pip run pip install --force-reinstall --no-deps torch torchvision --index-url https://download.pytorch.org/whl/cu128 afterwards. If an NVIDIA GPU sits unused, sift says so on start-up and prints
the command that fixes it. A machine without a GPU can borrow one.
sift add ~/handbook --name handbook # register a folder as a collection
sift index # read its files and build the keyword index
sift embed # embed every passage that has no vector yet
sift query "who approves travel expenses abroad"
sift serve # MCP server on http://127.0.0.1:8765/mcp, and the UI under /uisift index and sift embed are incremental: run them again after files change, and only what changed is processed.
An interrupted sift embed goes on where it stopped.
Start the server and leave it running: sift serve.
Register it with Claude Code, once, for all your projects:
claude mcp add --scope user --transport http sift http://127.0.0.1:8765/mcpKeep --scope user: without it, sift is registered for the current folder only, and sessions started anywhere
else don't see it, including those in the Claude desktop app's Code tab. Such a registration is removed, in the
folder where it was made, with claude mcp remove sift --scope local.
Start a new Claude Code session. claude mcp list shows sift as connected, and an agent asked for sift's status
lists your collections.
One server serves any number of agents, with the models loaded once. Other MCP clients, such as Cursor, VS Code or
Windsurf, connect the same way, to the streamable HTTP endpoint http://127.0.0.1:8765/mcp. To have every session
start a server of its own instead, over stdio: claude mcp add --scope user sift -- sift serve --stdio. Connectors
in claude.ai and in the Chat tab of the Claude apps can't use sift: claude.ai reaches connectors from Anthropic's
servers, which can't reach your computer.
Tool | Purpose |
| Exact keyword search, ranked by BM25: |
| Semantic search |
| Hybrid search for RAG: keyword and semantic candidates, reranked by a cross-encoder |
| Documents like a given one |
| A document's text by its id, all of it or only some |
| The collections: folders, counts, front-matter fields, links |
| A note's links, each with where it leads, and the notes linking to it |
| Link targets no note answers to yet |
| Documents that name a note without linking to it |
| Read files again right after writing, changing or deleting them; the one tool that changes anything: sift's index, never a file |
The search tools take k, collection, path, filters, fields, compact, all_passages, sort, links_to,
linked_from and adaptive, and search a time budget. Every result carries id (collection/path), title,
lines, score and the passage's text, so agents can cite their sources; compact: true leaves out the text, so an
agent can survey many results and read the ones that matter, and adaptive: true returns only those before the
largest drop in score, and a few more. Every parameter is described in the tools' schemas, and the server tells agents
on connecting which collections there are and which front-matter fields each has.
Command | Does |
| Register a folder as a collection, named after the folder by default ( |
| Read new and changed files, of every active collection or those named; with |
| Embed every passage without a vector, of every active collection or those named |
| Keyword, semantic and hybrid search |
| Documents like this one |
| A document, or some of its lines |
| Linked notes |
| Collections: folders, active or not, index files, sizes and counts, as JSON |
| Take collections into searches, or leave them out |
| Move collections' indexes to another folder or drive |
| Remove a collection and its index; its files stay on disk |
| Search, browse and read the index in your browser |
| The MCP server, and the UI under |
| Embed and rerank for other machines on this one's GPU (port 8766) |
| How well each search mode finds the answers you expect |
Search commands take -k, -c COLLECTION, --path GLOB, -f FIELD=VALUE filters, --fields, --compact,
--all-passages, --sort=-FIELD, --links-to, --linked-from, --adaptive and --json, and sift search
--budget SECONDS:
sift search '"notice period" AND (contract OR agreement)'
sift query "per diem abroad" -f departments=finance -f "created>=2024-06"
sift links vault/People/Hans.md --direction bothEvery command but sift move may run while sift serve does; the server finds what a command wrote with its next
search. docs/usage.md covers search syntax and scores,
filters and front matter, every format sift reads, linked notes, .siftignore, the UI, sift bench, and where the
indexes lie.
None needed: sift picks the device, the GPU memory budget and the batch size itself. Five optional environment variables exist for special cases:
Variable | Default | Purpose |
|
| The catalog, the logs and the worker's key, and the collections' indexes unless moved |
|
| Any Sentence Transformers embedding model |
|
| Any Sentence Transformers cross-encoder |
| none | A |
|
| Where scans are read: |
After changing the embedding model, run sift embed --reset. For business documents,
Snowflake/snowflake-arctic-embed-l-v2.0 ties the default on average: better on financial questions and long German
documents, worse on scientific claims, and 1.6 times as fast
(benchmarks).
A machine without a GPU can embed and rerank on another one's. Run sift worker there; it prints a line such as
SIFT_EMBED_REMOTE=https://<key>@workstation:8766 to set on the machine without. The texts travel over HTTPS, the
worker's certificate is pinned by the key, and the worker sees no index and no files. Setup, firewall and security:
docs/worker.md.
uv tool upgrade sift-rag # with pip: pip install -U sift-ragThen restart sift serve and sift worker, and start new agent sessions: an MCP client reads sift's tools when a
session starts, and a running session misses new ones. Machines that embed through a worker need the same major
version as the worker.
The index and the vectors carry over. When a release changes how the index is laid out, its first command updates it,
once, and says so, from what the index holds: no file is read again and nothing is embedded again, and an older sift
no longer opens it, so upgrade every sift that uses the index. From 1.7 or older, the index is split into a file per
collection: stop every other sift first, and have as much free space as the index takes (sift move FOLDER as the
first command splits it straight into that folder, on another drive too). The
changelog says what each release's update takes, such
as the next sift index building the keyword postings anew.
On one workstation (Core i9-14900K, RTX 4090), over an index of 45.7 million passages (arXiv, PubMed, OpenAlex and
more), searched warm through sift serve:
Use case | Mode | Time |
A name, an id, a term |
| 10 to 50 ms |
A question, by its keywords |
| 0.2 s |
By meaning |
| 0.6 s (0.1 s over 3.2 million) |
A question for RAG, reranked on the GPU |
| 0.6 s (0.3 s over 3.2 million) |
More documents like one |
| 0.7 s |
query scores nDCG@10 0.81 on scientific claims (SciFact), 0.69 on long German documents and 0.99 on German
questions, and ranks a known paper among 3.2 million at MRR@10 0.93. vsearch finds 99.75% of exact search's top 10
in a sixteenth of the memory. sift embed embeds 370 to 520 passages a second on the RTX 4090. Every figure, and how
it was measured: benchmarks.
Each collection's index is a SQLite file: documents, FTS5 keyword index and keyword postings, fp16 vectors and their
one-bit codes. Keyword search ranks by BM25 from the postings, skipping the ranges of passages that can't reach its
best results: milliseconds for a common word over 30 million passages, with FTS5's results. Semantic search compares a
query's code with every passage's, kept in RAM (96 bytes each), ranks the 10,000 nearest again by the query's own
values, and scores the best 1,000 exactly; it finds 99.75% of exact search's top 10 over 3.2 million abstracts in
about 0.1 seconds, with no index to build. query reranks the best 40 keyword and 40 semantic candidates on the GPU.
Embedding runs in batches of similar length, sized to the GPU memory left by other programs, and gives the GPU back
after 5 idle minutes. Details: docs/architecture.md.
In scope: text, Markdown and JSON Lines files, PDF (scans too), office documents, e-books, web pages and saved emails; Linux, macOS and Windows; NVIDIA GPUs (CUDA) and Apple GPUs (MPS), otherwise the CPU; one user, any number of local agents.
Out of scope for now: OCR of pictures that aren't PDFs; answer generation (sift retrieves, the agent writes);
approximate nearest-neighbour indexes; live file watching (an agent calls reindex after writing); remote access and
authentication; writing to your files.
sift follows semantic versioning. Within a major version, such as 1.x, these stay compatible: the commands, their options and their JSON output; the MCP tools and their parameters; the settings; the index, which every later 1.x opens and updates on first use if it must; and the worker protocol, so every 1.x embeds through every 1.x worker. sift's Python modules are its own and may change in any release.
git clone https://github.com/tobiaswuerth/sift
cd sift
uv venv && uv pip install --torch-backend auto -e ".[dev]" # or with pip: pip install -r requirements.txt -e ".[dev]"
pre-commit install # lint, format and config checks on every commit
pytest --cov # 100% line and branch coverage is requiredThe tests replace the models with small deterministic fakes, so they need no GPU and no downloads; everything else, the MCP server included, runs for real. See CONTRIBUTING.md for how changes and releases work.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Agentic search over your Dewey document collections from any MCP-compatible client.
Agent-driven search: build, import, tune, search, and score result quality — all over MCP.
Multi-engine search for AI agents. Trust scoring, local corpus, MCP-native. Self-hostable, BYOK.
Cloud or self-hosted knowledge for AI agents: hybrid search, reranking, GraphRAG, scoped MCP tools.
