Setup - Talqing

The server runs locally over stdio and talks to the Talqing API with your personal access token. Two environment variables configure it, both required.

Get the token and the command

Open the dashboard’s Agents page. At the top is a panel headed Build agents from your terminal, with a tab for Claude Code, Codex and Grok. It already carries your token and the base URL of the region you are currently looking at. The token is masked until you click Reveal token, and Copy always copies the real value — so the panel is safe to leave open on a shared screen. It is your MCP token, issued to you at signup and readable as often as you need it. It authenticates as you, acts with your current role, and reaches only your workspace. You can also use any personal access token you created yourself; see authentication.

Install

claude mcp add talqing \
  -e TALQING_BASE_URL=https://api.in.talqing.com \
  -e TALQING_API_KEY=$TALQING_API_KEY \
  -- npx -y @talqing/mcp

codex mcp add talqing \
  --env TALQING_BASE_URL=https://api.in.talqing.com \
  --env TALQING_API_KEY=$TALQING_API_KEY \
  -- npx -y @talqing/mcp

{
  "mcpServers": {
    "talqing": {
      "command": "npx",
      "args": ["-y", "@talqing/mcp"],
      "env": {
        "TALQING_BASE_URL": "https://api.in.talqing.com",
        "TALQING_API_KEY": "your-personal-access-token"
      }
    }
  }
}

[mcp_servers.talqing]
command = "npx"
args = ["-y", "@talqing/mcp"]
env = { TALQING_BASE_URL = "https://api.in.talqing.com", TALQING_API_KEY = "your-personal-access-token" }

{
  "mcpServers": {
    "talqing": {
      "command": "npx",
      "args": ["-y", "@talqing/mcp"],
      "env": {
        "TALQING_BASE_URL": "https://api.in.talqing.com",
        "TALQING_API_KEY": "your-personal-access-token"
      }
    }
  }
}

The Claude Code and Codex commands are run once in a terminal and write the client’s own configuration for you. Cursor’s block goes in .cursor/mcp.json in your project, or ~/.cursor/mcp.json to have it everywhere. Grok’s mcp add takes no --env flag, so its credentials can only be set in a file: ~/.grok/config.toml. The server needs Node 20 or newer. npx -y @talqing/mcp fetches the current release each time the client starts it; append @x.y.z to pin a version.

The two variables

Variable	What it is
`TALQING_BASE_URL`	The API base URL of one region — `https://api.in.talqing.com` or `https://api.us.talqing.com`
`TALQING_API_KEY`	Your personal access token, sent as `Authorization: Bearer <token>` on every call

Both are required and neither has a default. A missing one is a startup failure with the name of the variable, not a call that silently reaches somewhere else — there is no fallback host, so the server can never quietly talk to the wrong workspace.

TALQING_BASE_URL names a region. Everything this server can reach belongs to that one: an agent, its calls, its phone numbers and its credit balance. One token reaches every region — it names your workspace, not a place — so working in another region is the same token and a different base URL. The dashboard’s snippet always carries the region you are looking at; switch region there and copy it again. See regions.

Check that it works

Restart the client so it picks up the new server, then ask it something that only a working connection can answer:

List my Talqing agents.

It should call list_agents and come back with the agents in that region’s workspace, or say the workspace has none — either is a pass. A brand new workspace is empty, so if you want a reply with something in it, ask for the catalog instead:

Which LLM models can a Talqing voice agent use, and which of them do I have a provider key for?

That reads get_catalog and list_provider_keys, which exercises two operations and tells you something you need before you build anything. See provider keys.

Run a local install

npx downloads the package on every client start. To hold a fixed copy instead, install it globally — the package puts a talqing-mcp command on your PATH:

npm install -g @talqing/mcp

Then point the client at that binary rather than at npx:

{
  "mcpServers": {
    "talqing": {
      "command": "talqing-mcp",
      "args": [],
      "env": {
        "TALQING_BASE_URL": "https://api.in.talqing.com",
        "TALQING_API_KEY": "your-personal-access-token"
      }
    }
  }
}

Upgrading is then explicit: npm install -g @talqing/mcp@latest.

Troubleshooting

The server does not appear in the client

Check the client’s own MCP list first — claude mcp list for Claude Code, codex mcp list for Codex — to confirm the entry was written where you expected. Client configuration is usually per project as well as per user, so a server added in one directory is not there in another.If it is listed but not connecting, run the same command by hand with both variables set. The server fails at startup rather than mid-call, so a configuration problem prints its cause immediately:

TALQING_BASE_URL=https://api.in.talqing.com \
TALQING_API_KEY=$TALQING_API_KEY \
npx -y @talqing/mcp

A healthy start prints nothing and waits — it speaks MCP over stdin and stdout. Press Ctrl-C. A missing variable prints its name and exits. Anything else is Node: the package needs Node 20 or newer.

The token is rejected

A rejected token comes back as 401 with invalid token. Three causes, in order of likelihood:

The token was truncated on the way into the config. It is a long single-line string with no spaces and no prefix; copy it again with the dashboard’s Copy button rather than selecting the masked text.
The token was deleted. Revoking a personal access token deletes its row, and the row is the check — the token stops working on the next call, everywhere, immediately.
You are pointing at something that is not a region’s API. TALQING_BASE_URL must be https://api.in.talqing.com or https://api.us.talqing.com, with no path.

A 503 on the same call is different: it means the control plane could not be reached to verify the token. That is transient — retry.A 403 is not a token problem. The token works and your role does not allow that operation; see permissions.

Tools are missing, or the agent says it cannot do something

The server always advertises every operation regardless of your role — a viewer sees all of them and gets a 403 when they call a write. So a tool that is genuinely absent is one of two things.Some clients cap how many tools they will load from a server, or across all servers. This one exposes 123, which is enough to hit a cap that other servers never reach. Disable the MCP servers you are not using for this task and reconnect.Or the operation was never exposed. Sign-in, workspaces, members, roles, personal access tokens, buying credits, the streaming endpoints and deleting a call or its recording are deliberately unreachable, each for a stated reason — permissions lists them and, for the streams, the polling operation to use instead.

The client's context fills up

The tool list and the server instructions are sent on every model call, and 123 tools is a large fixed cost before your own conversation starts. The five biggest request types are already deferred rather than inlined — see tools — which is what keeps the surface loadable at all, but it is still the largest thing in the session.What helps: run one task per session rather than building an agent, debugging a call and writing your app in the same thread; keep other MCP servers off while you are working on Talqing; and let the agent fetch describe_schema for the one shape it is writing instead of asking it to dump all five.

Recipes

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

Tools

Every operation, and how its arguments map to the HTTP call.

​Get the token and the command

​Install

​The two variables

​Check that it works

​Run a local install

​Troubleshooting

​Next

Recipes

Tools

Get the token and the command

Install

The two variables

Check that it works

Run a local install

Troubleshooting

Next