ANHA in DeepSeek

DeepSeek has no connector screen of its own. You point an agent app at ANHA and choose DeepSeek as that agent's model.

Via an agentYour @handleAn agent appAbout 10 minutes

Install

Pick an agent that supports MCP
Any of the terminal guides here will do — OpenClaw, Hermes, Codex or Qwen Code.
Install ANHA in that agent by following its guide.
See the terminal guides
Choose DeepSeek as the model
In the agent's model settings, select DeepSeek and supply your API key.
The agent holds the ANHA connection; DeepSeek is only the model answering through it.
3
Writing your own program instead?
DeepSeek speaks OpenAI-compatible function calling. Load the ANHA tool definitions and pass them as functions.
Tool definitions
```
https://resolver.anha.ai/tools.json
```
Build your own agent

Run whichever list command your agent provides.

You should see

anha    connected    8 tools

A passing check only proves the connection works. This proves the model can actually reach the tools.

“Use ANHA to resolve @nike.ai and tell me what it can do.”

You cannot find an MCP setting in DeepSeek itself
Cause — There is not one. This is expected.
Fix — Connect ANHA to an agent app and select DeepSeek as its model.
The tools are listed, but every call answers that you are not signed in
Cause — The browser approval was closed before it finished, or you approved with an email address that owns no handle.
Fix — Run npx -y anha-mcp login again and sign in with the same email you registered the handle under. If you have no handle yet, register one first — the approval screen has nothing to offer otherwise.
An order or payment just hangs
Cause — This is not a fault. The merchant requires phone approval, so the call returns a pending status with a request id.
Fix — Approve it on your phone. A human decision can take minutes, and the caller polls until it lands.

Every guide shares the same sign-in and the same tools, so a fix on another platform often applies here too.