Overview
How It Works
Notery provides two tools that AI coding assistants can use to query your workspace documentation: notery_search and notery_get. These tools are available through the MCP server and agent skills.
The typical workflow is a two-step search-then-retrieve pattern:
Step 1: Search
Your AI assistant uses notery_search with a natural language query. Notery performs semantic search using Cohere embeddings and pgvector, returning a ranked list of matching documents with titles, paths, IDs, similarity scores, and summaries.
You: "Search my docs for how authentication works"
notery_search → returns 5 results with summaries and IDs
Step 2: Retrieve
Your assistant reviews the search results, picks the most relevant document, and uses notery_get with that document's ID to fetch the full content.
notery_get(id: "550e8400-...") → returns complete document text
This two-step pattern keeps responses fast. The search step returns lightweight summaries so your assistant can identify what's relevant, then full content is only fetched for the documents you actually need.
Why Two Tools?
A single "search and return everything" tool would be slow and wasteful. Most workspaces have documents with hundreds or thousands of words. Returning full content for 10 search results would flood your assistant's context window with irrelevant text.
The search → get pattern lets your assistant:
- Scan broadly — review summaries of many documents quickly
- Read selectively — fetch full content only for what matters
- Combine sources — pull in content from multiple documents when a question spans several topics
Connection Methods
There are two ways to connect your AI tools to Notery:
| Method | How it works | Best for |
|---|---|---|
| MCP Server | Your tool connects to https://notery.dev/api/mcp over HTTP | Tools with MCP support (Claude Code, Cursor, VS Code) |
| Agent Skills | Download a skills package into your project | Tools without MCP support |
Both methods provide the same notery_search and notery_get tools. The MCP server requires no downloads — your tool connects directly. Agent skills are a self-contained fallback.
See Installation for setup instructions, or read the individual tool docs:
- notery-search — semantic search across your workspace
- notery-get — retrieve full document content by ID
- MCP Server — technical details of the MCP endpoint