接続Connect your AI

Connect DeepSeek

The DeepSeek chat app has no connector settings. DeepSeek’s own agent harness, dsh, is an MCP client.

Steps

  1. Create an Adako API key at /keys.
  2. Add the block below to your dsh config, or to a patch file you pass to dsh.
    dsh.yml
    - insert:
        - id: mcp-adako
          name: '@deepseek-ai/dsh-mcp-client'
          config:
            serverName: adako
            transport: streamable-http
            url: 'https://adako.ai/mcp'
            headers:
              Authorization: 'Bearer ak_live_…'
  3. Start the harness with npx @deepseek-ai/dsh web. It reads Adako’s tools on startup.
  4. Ask it to use Adako.
Untested, and dsh is a developer preview whose configuration still changes. Its MCP plugin documents static headers only, not the browser sign-in, so use an API key. Neither the DeepSeek chat app nor the DeepSeek API can reach an MCP server on its own: to drive Adako with DeepSeek models, run them inside a client from this list.

Test it

Ask for something read-only first. This call is free and changes nothing.

first prompt
Using Adako, tell me which ad accounts are connected.

You should get back what is connected, which account is active, and two or three prompts worth trying next. If nothing is connected yet, the answer says so and links to the Accounts page.

When a write comes back as a proposal

Under the inbox policy, a write returns a proposal id and a diff instead of a result. Say “approve it” in the chat, or open Approvals and approve it there. Under direct, your confirmation in the chat is the approval.

Last updated