@talqing/mcp is a local stdio MCP server that lets Claude Code, Codex, Cursor or any MCP client build, publish and operate Talqing agents. Use it when you want to build an agent in the same session as the code its tools call, or ask questions of real call data from your terminal. It gives your client two things:
  • One tool per API endpoint, named after the endpoint’s operation (create_agent, publish_agent, get_call). Arguments mirror the HTTP call. See tools.
  • Server instructions — the same platform guide the in-product CoPilot is prompted with: how an agent config fits together, draft and publish, the operation tree, channels and model selection. Your client receives it automatically on connect.
Both are generated from the registry the CoPilots use, so anything a CoPilot can do, your coding agent can do.

​
What it covers

Agents, tools, agent tasks, FAQs, integrations and triggers, carrier accounts, phone numbers, media streams, secrets, provider keys, webhooks, the catalog; placing outbound calls; running call, email and WhatsApp batches; opening a text chat and talking to an agent; and reading back calls, conversations, transcripts, cost and latency. A few areas are deliberately not exposed — sign-in, members, roles, tokens, buying credits, streams, and deleting call records. See permissions.

​
It runs as you

Every call uses your personal access token: your workspace, your current role, the same result curl would give. A viewer can read everything and gets a 403 on writes. The server adds no permissions and no service account.

​
Next

Quickstart

One command and four prompts to a published agent.

Setup

Configuration for every client.

Recipes

Prompts that build, run and debug end to end.