# Pydantic AI Demo
This directory lets you run the Mem0 MCP server end-to-end with the bundled
Pydantic AI agent without touching `uvx` or Smithery.
## Requirements
```bash
pip install -e ".[smithery]" # or uv pip install -e ".[smithery]"
export MEM0_API_KEY="sk_mem0_..."
export OPENAI_API_KEY="sk-openai_..."
```
## Run the REPL
```bash
python example/pydantic_ai_repl.py
```
What happens:
1. The script loads `example/config.json`, which points to your local
`mem0_mcp_server.server` module (`python -m mem0_mcp_server.server`).
2. The Mem0 MCP server starts over stdio with whatever env vars you exported.
3. A Pydantic AI agent (Mem0Guide) connects to that server and drops you into a
REPL. Type prompts like “remember that I love tiramisu” or “search memories
for tiramisu” to exercise the tools.
### Custom configs
- Set `MEM0_MCP_CONFIG_PATH=/path/to/another/config.json` to point the agent to
a different MCP endpoint (e.g., your Smithery deployment or Docker container).
- Use `MEM0_MCP_CONFIG_SERVER` if the config file defines multiple servers.
This flow mirrors the original bundled script but lives entirely in this
`example/` directory with ready-to-use config so the REPL “just works” for local
testing.
## Configuring Other Transports
The two config files in this folder cover the most common setups:
- `config.json` – launches your local checkout via stdio (`python -m mem0_mcp_server.server`).
- `docker-config.json` – points at an HTTP server on `http://localhost:8081/mcp` (e.g., the Docker container).
You can copy either file and tweak the `command`/`args` or `type`/`url` fields
to match other environments—or reuse them directly in other MCP clients (Claude,
Cursor, custom apps). The Pydantic REPL is just a convenient example of how any
client would connect once it has the right config.
- **Local stdio (default)** – runs `python -m mem0_mcp_server.server`. No edits
needed; the REPL launches the server from your working tree.
- **uvx / published package** – change `command` to `"uvx"` and `args` to
`["mem0-mcp-server"]` if you want to test the installed CLI instead of your
repo checkout.
- **Docker / hosted HTTP** – expose the container on `http://localhost:8081/mcp`
(or your host URL) and create a config that uses Pydantic’s HTTP transport,
e.g. `example/docker-config.json`. Set `MEM0_MCP_CONFIG_PATH` to that file
before running the REPL.
- **Smithery-hosted HTTP** – once deployed, copy the Smithery URL
(`https://server.smithery.ai/.../mcp`) into the same HTTP config shape above
so the REPL talks to the hosted server instead of starting a local process.
In all cases you can select the desired server definition with
`MEM0_MCP_CONFIG_SERVER` if your config file contains multiple entries.
