From zero to a searchable, agent-ready index in ten minutes.

1. 1

   Install StrataFS

   Install the single StrataFS binary with your package manager — npm, pip, Homebrew, curl, Docker, or from source. The same binary ships the indexer, the REST API, and the MCP server.

   npmpipcurlHomebrewDockerSource

   npm install -g stratafs
   stratafs config init
   stratafs serve

   macOS · Linux · WindowsCopy

   Prefer the long version? See all install methods →

2. 2

   Initialise a config

   Run stratafs config init to write a starter config.toml with a sensible default: your current directory indexed as a local, read-only source.

3. 3

   Add a storage source

   Point StrataFS at what you want to search. A local folder works out of the box; add an S3 bucket, GCS, Azure, SharePoint, Google Drive, or Jira project with a few lines of read-only credentials.

   # ~/.stratafs/config.toml
   [[sources]]
   name = "docs"
   type = "local"
   path = "/Users/you/work/docs"
   
   [[sources]]
   name = "s3-archive"
   type = "s3"
   bucket = "acme-knowledge"
   region = "eu-west-1"
   # read-only IAM credentials from the environment

4. 4

   Start the server

   Run stratafs serve. StrataFS indexes on first run — parse, chunk, and embed locally — then brings up the REST API on :8080 and the MCP server on :8081.

   $ stratafs serve
   ✓ indexed 12,481 files across 2 sources (41s)
   ✓ REST API   http://localhost:8080
   ✓ MCP server http://localhost:8081/mcp

5. 5

   Run your first hybrid search

   Query from the CLI or the REST API. One search fuses FTS5 BM25, vector cosine similarity, and metadata into a single ranked result set — no separate vector store to manage.

   $ strata search "how do we rotate the signing keys" --top 3
   1. ops/runbooks/key-rotation.md      score 0.94
   2. docs/security/secrets.mdx          score 0.81
   3. infra/terraform/kms.tf             score 0.76

6. 6

   Connect an AI agent over MCP

   Point Claude Desktop, Cursor, or any MCP client at http://localhost:8081/mcp. The agent discovers stratafs.search and answers from your actual files, offline.

   {
     "mcpServers": {
       "stratafs": { "url": "http://localhost:8081/mcp" }
     }
   }

Where to go next

• Architecture — how the pipeline, storage, and index fit together.
• The search engine — the hybrid BM25 + vector + metadata query, explained.
• Storage backends — config for S3, GCS, Azure, SharePoint, Drive, and Jira.
• Glossary — hybrid search, BM25, embeddings, MCP, and more in one place.
• MCP server — connect Claude Desktop, Cursor, and custom agents.