MCP server - Talqing

Talqing ships an MCP server, @talqing/mcp. Point Claude Code, Codex or any other MCP client at it and your coding agent builds, publishes and operates agents on Talqing without leaving the terminal. The server hands your client two things. The first is one tool per API endpoint — 123 of them, covering everything a workspace is built out of. The second is the same platform knowledge our own in-product CoPilot is prompted with: how an agent is put together, draft and published, the operation tree, lifecycle hooks, channels and model selection. Your client receives that automatically as server instructions, so anything the CoPilot can build, your coding agent can build. Both are generated from the same registry the CoPilot binds its own tools from, so the two surfaces cannot drift apart. An endpoint that exists is a tool; a rule the CoPilot follows is a rule your client is told.

What you get

Area	What your agent can do
Agents	Create, read, update and delete an agent; validate it; publish it; read a past version and roll back to it
Agent tasks	Write a task’s inputs, output shape and model stack; run it; read the run’s output, trace, steps and cost
Tools	Author a tool, its `json_schema` and its operation tree; run the draft for real; validate; publish; roll back
Knowledge bases	Crawl a site, choose which pages to index, build, poll to `ready`, fix an entry by hand, regenerate the table of contents
Integrations and triggers	Connect a provider, list what its MCP server exposes, bind an inbound event to a text agent
Carrier accounts	Read what each carrier requires, record credentials, provision the SIP trunk, list the numbers the account owns
Phone numbers	Import, provision, assign a published agent, unassign, change per-number settings, read `readiness`
Secrets, provider keys and webhooks	Store a workspace secret; add or rotate a BYOK provider key; subscribe to events, rotate a signing secret, replay deliveries
Running agents	Place one outbound call; create and steer a call batch or an email batch; open a text conversation and talk to an agent; mint the token a browser needs to join a web call
Reading it back	Calls with transcript, every tool call and its output, per-stage latency and cost; conversations and the sessions inside them; workspace volume, spend and response time per day
Catalog	Every valid `provider`/`model` pair with what it supports, the voice and avatar galleries, an ElevenLabs voice’s own tuning

The full list, with the HTTP call each one maps to, is on tools.

Who it is for

You already work inside a coding agent, and want the voice agent written in the same session as the API its tools will call.
You are testing changes against a real workspace and would rather say what you want than assemble a config by hand.
You are debugging: read the last five failed calls and tell me why is a question the server can answer, because get_call returns the transcript, the tool calls and the per-stage latency of a call that already happened.

It is your API, with your permissions

The server adds nothing and hides nothing. Every call it makes is a call you could make yourself with the same token: the same operations, the same workspace, the same role. A view-only member gets the read operations and a clear 403 on the rest, exactly as in the dashboard. Four areas are out of reach by design, and permissions says which and why. TALQING_BASE_URL names a region, and everything the server can reach belongs to that one — an agent, its calls, its phone numbers and its credit balance. One token reaches every region, so working in another one is the same token and a different base URL. If you want to see it work before reading further, the coding-agent quickstart is four prompts from an empty workspace to a published agent with a tool attached.

Setup

Install and configure it for your client, and verify it works.

Recipes

Prompts that produce a published agent, a live phone line and a campaign.

Tools

Every operation, how arguments map to HTTP, and the deferred schemas.

​What you get

​Who it is for

​It is your API, with your permissions

​Next