Cowpin Agent Memory MCP Server
Official# @cowpin/mcp-server
Official [Model Context Protocol](https://modelcontextprotocol.io/) stdio bridge for **[Cowpin](https://www.cowpin.com)** — persistent bookmark memory, URL archiving, and semantic search for **Cursor**, **Claude Desktop**, and other MCP clients.
[](https://glama.ai/mcp/servers/cowpin-ai/mcp-server)
**Setup (Cursor):** [Connect Cowpin to Cursor](https://www.cowpin.com/docs/agent/connect-cursor)
## Install
```bash
npx -y @cowpin/mcp-server@latest
```
## Verify (offline — for directory reviewers)
Open-source stdio bridge with **no hosted secret**: runs JSON-RPC lines against a mock gateway.
```bash
npx -y @cowpin/mcp-server --self-test
```
From a git checkout of [cowpin-ai/mcp-server](https://github.com/cowpin-ai/mcp-server):
```bash
npm test
```
Tool calls against production require a `cw_agent_…` key ([crypto auth](https://www.cowpin.com/docs/agent/crypto-auth)); the bridge itself is fully inspectable in this repository.
## Environment
| Variable | Required | Description |
| :--- | :--- | :--- |
| `COWPIN_AGENT_API_KEY` | Yes (for tools) | `cw_agent_…` from [wallet auth](https://www.cowpin.com/docs/agent/crypto-auth) |
| `COWPIN_API_URL` | No | Default `https://www.cowpin.com/api/agent/mcp` |
## Cursor `.cursor/mcp.json`
```json
{
"mcpServers": {
"cowpin": {
"command": "npx",
"args": ["-y", "@cowpin/mcp-server@latest"],
"env": {
"COWPIN_AGENT_API_KEY": "cw_agent_your_key_here"
}
}
}
}
```
## Smithery
Registry config: `smithery.json` in this package. Listing: [smithery.ai](https://smithery.ai) — search `@cowpin/mcp-server`.
## License
MIT
TDQS
Scored across 4 tools
Each tool has a clearly distinct purpose: saving, searching, checking balance, and retrieving full memory pages. Even though search returns snippets and get_memory_page returns full text, they serve different functions and are unlikely to be confused.
All tools follow the consistent pattern cowpin_<verb>_<object> (save_memory, search_memory, get_balance, get_memory_page). The 'get' prefix is used consistently for retrieval operations, making the naming predictable and uniform.
With 4 tools, the server is well-scoped for its purpose as a memory vault. The count is neither too thin nor excessive, and each tool contributes to core workflows (save, search, retrieve, and billing awareness).
The server covers save, search, and full-text retrieval, which are the primary operations for a memory vault. However, it lacks update and delete capabilities, which could be relevant if users need to correct or remove stored memories, though this may be intentional.