@andrewkimjoseph/celina 0.2.5 → 0.2.6

package/README.md

@@ -102,6 +102,98 @@ For local write tools, add a funded mainnet wallet:
 
 Never commit private keys. Use env vars only.
 
+## Local LLM integration
+
+Celina is an **MCP tool server**. A local LLM stack needs an **MCP client** that can connect to Celina and pass tool definitions to a model that supports **function / tool calling**.
+
+Read-only tools (balances, blocks, GoodDollar status, etc.) work out of the box. For local write tools (`send_token`, `estimate_send`), set `CELO_PRIVATE_KEY` in the MCP server `env` block (stdio) or use the [hosted encryption flow](#write-tools-hosted-mode) (HTTP).
+
+### LM Studio (0.3.17+)
+
+LM Studio can host MCP servers directly via `mcp.json` (same format as Cursor).
+
+1. Open LM Studio → **Program** → **Install** → **Edit mcp.json**
+2. Add Celina under `mcpServers`
+3. In **Server Settings**, enable **Allow calling servers from mcp.json**
+4. Chat with a tool-capable model (e.g. Qwen 2.5, Llama 3.1+)
+
+```json
+{
+  "mcpServers": {
+    "celina": {
+      "command": "npx",
+      "args": ["-y", "@andrewkimjoseph/celina"],
+      "env": {
+        "CELO_PRIVATE_KEY": "0x..."
+      }
+    }
+  }
+}
+```
+
+Omit `CELO_PRIVATE_KEY` if you only need read-only chain queries.
+
+### Open WebUI + Ollama
+
+[Open WebUI](https://docs.openwebui.com/features/extensibility/mcp/) supports **streamable HTTP** MCP natively (not stdio).
+
+**Hosted Celina (easiest):** Admin Settings → External Tools → **Add Server** → Type: **MCP (Streamable HTTP)** → URL:
+
+```
+https://mcp.celina.andrewkimjoseph.com/mcp
+```
+
+**Local HTTP server:** run Celina in HTTP mode, then point Open WebUI at it:
+
+```bash
+npm run build
+npm run start:http
+```
+
+Add an External Tool with Type **MCP (Streamable HTTP)** and URL `http://localhost:10000/mcp`.
+
+For write tools over HTTP, set `WALLET_ENCRYPTION_PRIVATE_KEY` in `.env` (see [Deploy to Render](#deploy-to-render)) and use the [encrypt-key flow](#write-tools-hosted-mode).
+
+> If Open WebUI runs in Docker, use `http://host.docker.internal:10000/mcp` instead of `localhost`.
+
+### Continue (VS Code)
+
+[Continue](https://docs.continue.dev/customize/deep-dives/mcp) works with local models (Ollama, LM Studio, etc.) in **agent mode**.
+
+Create `.continue/mcpServers/celina.yaml` in your workspace:
+
+```yaml
+name: Celina
+version: 0.0.1
+schema: v1
+mcpServers:
+  - name: celina
+    type: stdio
+    command: npx
+    args:
+      - "-y"
+      - "@andrewkimjoseph/celina"
+    env:
+      CELO_PRIVATE_KEY: "0x..."
+```
+
+Alternatively, copy the [local stdio JSON](#local-stdio-npm) from the Cursor section into `.continue/mcpServers/mcp.json` — Continue picks up Claude/Cursor-style configs automatically.
+
+### Test without an LLM
+
+Use MCP Inspector to call Celina tools directly over stdio:
+
+```bash
+npm run build
+npm run inspect
+```
+
+### Tips
+
+- Use models with reliable tool-calling support; small or older models may skip tools or call them incorrectly.
+- Start with read-only prompts, e.g. *"What's the USDm balance of 0x…?"* or *"Is this wallet GoodDollar whitelisted?"*
+- Keep private keys in env vars only — never commit them to config files in git.
+
 ## Write tools (hosted mode)
 
 Write tools (`send_token`, `estimate_send`) accept an RSA-encrypted private key per request — never plaintext.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@andrewkimjoseph/celina",
-  "version": "0.2.5",
+  "version": "0.2.6",
   "description": "Celina — MCP server for Celo mainnet. Balances, transfers, and chain reads for LLM agents.",
   "type": "module",
   "main": "build/index.js",