@squadrules/mcp 5.0.0-beta.1 → 5.0.0

package/README.md

@@ -102,7 +102,7 @@ The current codebase includes:
 
 - **HTTP application server** — Express app for MCP, REST, auth routes, and UI
 - **stdio MCP transport** — direct local-host launch path for desktop/IDE MCP clients
-- **Qdrant-backed adapter store** — required for runtime
+- **Vector/trace store** — embedded LanceDB by default; Qdrant when `QDRANT_URL` is set
 - **Optional Redis cache / proof-of-work state store** — enabled when `REDIS_URL` is set
 - **Optional Keycloak auth integration** — browser session + Bearer JWT validation
 - **React UI** — served from the same origin at `/ui`
@@ -134,15 +134,13 @@ Compose — install the package globally and point your host at `squadrules serv
 ### Prerequisites
 
 - **Node.js 24+**.
-- **A Qdrant instance on `http://localhost:6333`** — SquadRules cannot start
-  without it, and no auth is required for local use. If you don't already run
-  one, this is the quickest option (optional convenience):
-
-  ```bash
-  docker run -p 6333:6333 qdrant/qdrant
-  ```
-
-- **One embedding backend**, supplied through the host `env` below.
+- **One embedding backend**, supplied through the host `env` below (the embedded
+  store removes the vector *database* dependency, not the embedding provider).
+- **No database required.** By default SquadRules keeps its vectors in an embedded,
+  file-backed LanceDB store under `~/.config/squadrules/lancedb` (created on first
+  run) — no Qdrant, Redis, or Docker. To keep an existing Qdrant, set `QDRANT_URL`;
+  the single-switch selection rule and limitations live in
+  [Known issues & limitations](docs/known-issues-and-limitations.md#vector-store-backends).
 
 ### Install
 
@@ -167,23 +165,24 @@ Add SquadRules to your host's `mcp.json` (Cursor, Claude Desktop, Claude Code,
     "SquadRules": {
       "command": "squadrules",
       "args": ["serve"],
-      "env": {
-        "QDRANT_URL": "http://localhost:6333",
-        "QDRANT_API_KEY": "",
-        "OPENAI_API_KEY": "sk-..."
-      }
+      "env": {}
     }
   }
 }
 ```
 
-`QDRANT_API_KEY=""` selects no-auth localhost Qdrant. For the embedding backend,
-supply **one** of:
+An empty `env` is enough: with no `QDRANT_URL` the server uses the embedded
+LanceDB store (created on first run under `~/.config/squadrules/lancedb`), and
+embeddings run **locally with fastembed** (no API key, no inference service; the
+model is downloaded once into `~/.config/squadrules/models`). To use an existing
+Qdrant instead, add `"QDRANT_URL": "http://localhost:6333"` (and
+`"QDRANT_API_KEY": ""` for no-auth localhost Qdrant). To override the local
+embedding default with an external backend, supply **one** of:
 
 - **OpenAI** — `OPENAI_API_KEY`
 - **Ollama / OpenAI-compatible** — `OPENAI_API_URL`, `OPENAI_EMBEDDING_MODEL`,
   and `OPENAI_API_KEY=ollama`
-- **TEI** — `TEI_BASE_URL` (+ optional `TEI_MODEL`)
+- **TEI** (deprecated) — `TEI_BASE_URL` (+ optional `TEI_MODEL`)
 
 Every parameter is **ENV-overridable**. To run SquadRules as an HTTP listener
 instead of stdio, add `"--transport", "http"` to `args` (see
@@ -311,11 +310,15 @@ health endpoint (`curl http://localhost:3000/health`).
 
 ### Embeddings fail on startup
 
-Set one working embedding backend in the host `env`:
+By default embeddings run **locally with fastembed** — no key required — but the
+**first run downloads the model**, so it needs network access once (cached under
+`~/.config/squadrules/models`; air-gapped hosts must pre-seed it). Override the
+model or location with `FASTEMBED_MODEL` / `FASTEMBED_CACHE_DIR`. To use an
+external backend instead, set one working provider in the host `env`:
 
 - OpenAI: `OPENAI_API_KEY`
 - Ollama/OpenAI-compatible: `OPENAI_API_URL`, `OPENAI_EMBEDDING_MODEL`, `OPENAI_API_KEY=ollama`
-- TEI: `TEI_BASE_URL` (+ optional `TEI_MODEL`)
+- TEI (deprecated): `TEI_BASE_URL` (+ optional `TEI_MODEL`)
 
 ### The CLI keeps asking for login