Provides RAG-based search over an Obsidian vault, allowing AI agents to retrieve and answer questions from notes with source citations.
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., "@obsidian-rag-mcp search my notes for information on the CIA triad"
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.
Fully local RAG over an Obsidian (or any markdown) vault, powered by LM Studio. Nothing leaves your machine.
Two ways to use it:
obsidian-rag — a terminal REPL: ask questions, get answers grounded in
your notes with source citations.
obsidian-rag-mcp — an MCP server for the LM Studio GUI: the chat model
gets a search_notes tool and answers from your vault, inside the app.
Indexing is incremental: only new or edited files are re-embedded on each run.
LM Studio with the local server running
(Developer tab → Start Server, default http://localhost:1234).
Two models loaded:
a chat model (e.g. any Gemma / Llama / Qwen instruct model)
an embedding model (e.g. nomic-embed-text-v1.5)
Python ≥ 3.11 and uv (or pipx).
Related MCP server: obsidian-local-mcp
uv tool install git+https://github.com/shirokoweb/rag-obsidian-lmstudio
# or: pipx install git+https://github.com/shirokoweb/rag-obsidian-lmstudioobsidian-rag --docs-dir ~/path/to/your/vaultChat: google/gemma-4-e4b
Embed: text-embedding-nomic-embed-text-v1.5
Indexed 317 chunks from /Users/you/vault
Ask a question (blank line or Ctrl-D to quit).
? What is the CIA triad?
The CIA triad is a model that helps organizations consider risk ...
sources: Module 2/25. Explore the CIA triad.md, ...
top score: 0.830Add the server to LM Studio's mcp.json (Program tab → Install →
Edit mcp.json):
{
"mcpServers": {
"obsidian-rag": {
"command": "obsidian-rag-mcp",
"env": {
"RAG_DOCS_DIR": "/Users/you/path/to/your/vault"
}
}
}
}Then ask the chat model anything about your notes — it calls search_notes
and answers grounded, with source filenames.
If LM Studio can't find the command, use the absolute path (
which obsidian-rag-mcp) in thecommandfield.
CLI flags take precedence over environment variables.
Flag | Env var | Default | Purpose |
|
| (required) | Vault / notes directory |
|
|
| LM Studio server URL |
|
| auto-detect | Chat model id |
|
| auto-detect | Embedding model id |
|
|
| Retrieved chunks per question (1–20) |
Auto-detection picks the first loaded model whose id contains embed as the
embedder and the first other model for chat. With several chat models loaded,
set RAG_CHAT_MODEL explicitly.
The embedding cache lives in your OS user-cache directory (e.g.
~/Library/Caches/rag-obsidian-lmstudio on macOS) — never inside your vault.
Deleting it is always safe; it will be rebuilt.
Symptom | Fix |
| Start the server: LM Studio → Developer tab → Start Server |
| Load an embedding model and a chat model in LM Studio |
| The chat model is too large/slow — set |
| Pass |
Stale answers after editing notes | Nothing to do — the index refreshes on every run/query |
All traffic goes to your configured LM Studio URL (localhost by default); there are no other network calls, no telemetry.
The cache uses plain JSON + NumPy .npz — no pickle, nothing executable.
The tool only ever reads your vault; it never writes into it.
git clone https://github.com/shirokoweb/rag-obsidian-lmstudio
cd rag-obsidian-lmstudio
uv sync --all-extras
uv run pytest # tests
uv run ruff check . && uv run ruff format --check .
uv run mypy src # typecheckSee SPEC.md for design decisions. MIT license.
Search the user's markdown notes for passages relevant to the query. Use this for any question about the user's notes, then answer using only the returned passages and cite their source filenames.
Args: query: A natural-language question or topic to search for. top_k: How many passages to return (default 4, max 20).
| Name | Required | Description | Default |
|---|---|---|---|
| query | Yes | ||
| top_k | No |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations exist, so the description carries the transparency burden. It clearly implies a read-only search operation and discloses that results are passages with source filenames. However, it does not mention edge-case behavior (e.g., no results or duplicated passages), which is a minor omission for a search tool.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is concise, front-loaded with the primary purpose, and uses a clear 'Args' block for parameter details. Every sentence earns its place with no redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given a simple 2-parameter search tool and the presence of an output schema, the description covers purpose, usage, and parameters comprehensively. It even includes post-condition instructions (cite sources), making it complete for typical use.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema descriptions are absent (0% coverage), but the description fully compensates by explaining 'query' as a natural-language question and 'top_k' as a count of passages with default and max values. This adds necessary semantic meaning beyond the raw schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific action ('Search the user's markdown notes for passages relevant to the query') and reinforces its role as the tool for any note-related question. The verb and resource are unambiguous, making the tool's purpose immediately clear.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explicitly instructs when to use this tool ('Use this for any question about the user's notes') and how to handle the output ('answer using only the returned passages and cite their source filenames'). This provides concrete usage context, even without listing alternative tools.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.
v0.1.1search_notesWith only one tool, there is no possibility of confusion between tools. The single tool's purpose is clear and distinct.
The tool name 'search_notes' follows a clear verb_noun pattern. As the only tool, it is internally consistent.
A single tool is borderline. It is not trivial and serves the core purpose of searching notes, but the tool surface feels thin for a server intended to support RAG over an entire note collection.
The search tool covers the primary retrieval need. Minor gaps exist, such as no ability to list all notes or fetch a specific note directly, but these can be worked around by searching with broad queries.
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
Serve a folder of Markdown notes as an MCP server: hybrid search, reading, and sourced answers.
Markdown-based note-taking with a hosted MCP server. Your notes serve you and your AI.
Google Keep-style notes app with an MCP server for AI agents to read/write notes.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/shirokoweb/rag-obsidian-lmstudio'
If you have feedback or need assistance with the MCP directory API, please join our Discord server
