dots-mcp
# dots-mcp
> An MCP server for **OpenAI Dots** — the always-on, proactive agents announced at DevDay 2026.
_Unofficial. Not affiliated with or endorsed by OpenAI._
## Status
OpenAI has not published a public Dots API yet. Until it does, `dots-mcp` gives any MCP client (Claude, Cursor, ChatGPT, etc.) live, searchable access to the community-curated [awesome-dots](https://github.com/mergisi/awesome-dots) list. Dots-native tools will be added here as soon as an API is available.
## Tools
| Tool | Description |
| --- | --- |
| `list_dots_resources` | List resources, optionally filtered by section (`Official`, `News`, `Tools`…). |
| `search_dots_resources` | Keyword search across names, descriptions and sections. |
The list is fetched from GitHub and cached for 10 minutes. Override the source with `DOTS_LIST_URL`.
## Install
```bash
git clone https://github.com/mergisi/dots-mcp.git
cd dots-mcp
npm install && npm run build
```
### Claude Code
```bash
claude mcp add dots -- node /absolute/path/to/dots-mcp/dist/index.js
```
### Claude Desktop / Cursor
```json
{
"mcpServers": {
"dots": {
"command": "node",
"args": ["/absolute/path/to/dots-mcp/dist/index.js"]
}
}
}
```
## Roadmap
- [x] Search the awesome-dots resource list
- [ ] Dots-native tools (send a task, check status, list Dots) once OpenAI ships a public API
- [ ] npm package (`npx dots-mcp`)
## Contributing
Issues and PRs welcome. To add a resource to the list, open a PR on [awesome-dots](https://github.com/mergisi/awesome-dots).
## License
MIT
TDQS
Scored across 2 tools
list_dots_resources and search_dots_resources both retrieve from the same dataset, but the list vs. keyword-search split is a clear, conventional distinction. An agent can easily choose the right tool based on whether it wants to enumerate/filter by section or query by keyword.
Both names use consistent snake_case with a verb_noun pattern (list_dots_resources, search_dots_resources) and share the same noun phrase. There is no deviation or mixed convention.
Two tools is on the thin side for a resource-access server; while both are warranted, the surface feels minimal. A third tool (e.g., get resource details) could round it out, but the current count is not unreasonable for a narrow, read-only list.
The domain is a curated resource list about OpenAI Dots; listing (with optional section filter) and keyword search cover the primary retrieval needs. However, there is no explicit operation to fetch a single resource by ID or to list available sections, leaving minor gaps an agent may need to work around.