The @talqing/mcp server gives Claude Code, Codex or any MCP client one tool per Talqing API endpoint, plus the platform knowledge the in-product CoPilots use. This page takes you from nothing to a published agent in four prompts. For other clients and configuration options, see MCP setup.
1

Copy the command from the dashboard

Open Agents in the dashboard. The Build agents from your terminal panel has a tab per client with your token and the API base URL already filled in. Click Copy — it copies the real token even while it is masked.The token is your personal access token: it acts as you, with your current role, in this workspace only.
2

Run it once

Claude Code
For Codex, the same command is codex mcp add talqing --env … -- npx -y @talqing/mcp. The server needs Node 20 or newer.
3

Check the connection

Restart the client and ask:
Which LLM models can a Talqing voice agent use, and which of them do I have a provider key for?
It should call get_catalog and list_provider_keys. If it cannot, see troubleshooting.

​
Build an agent in four prompts

Each prompt is a plain request in your client’s chat.
  1. Create the agent.
    Create a voice agent called “Northwind support” that answers questions about orders, delivery and returns for an online hardware store. Read the catalog first and pick a model stack I have keys for. Do not publish it.
    It reads the catalog and your provider keys, writes a prompt and greeting, and calls create_agent.
  2. Add a tool.
    Add a tool called lookup_order that takes an order number, calls GET https://api.northwind.example/orders/{{args.order_number}} with an Authorization header from the workspace secret NORTHWIND_API_KEY, and returns the status and delivery date. Attach it to the agent.
    It writes the tool, runs validate_tool and publish_tool (a tool must be published before an agent can attach it), then adds it to the agent’s tools list. The secret must already exist — see secrets.
  3. Validate.
    Validate the agent and paste the result.
    validate_agent checks what a draft cannot show: missing provider keys, unpublished tools, unsupported models. Errors block publishing; warnings do not. See validation.
  4. Publish.
    Publish it.
    publish_agent freezes the draft as a version and pins each attached tool. Test it from the agent’s page in the dashboard, or give it a phone number.

​
What it will not do unasked

The server’s instructions tell the model to never publish an agent, never call or message a real person, and never delete anything unless you ask for that specific thing. These are instructions, not enforcement — keep your client’s approval prompts on. Permissions lists what the server cannot reach at all and which operations to approve by hand.

​
Next

MCP recipes

Prompts for phone numbers, campaigns, agent tasks and debugging.

MCP setup

Cursor, Grok, generic stdio clients and a pinned install.