MCP troubleshooting
The messages you may see when an assistant works with Telenow, what they mean, and what to do.
Connecting
| What you see | What to do |
|---|---|
| The Telenow page offers no organization to connect | Only an owner or admin can connect an organization. Ask one to connect, or to change your role. |
| "This organization has no phone number yet" | Get a number (Phone numbers), then reload the connect page. |
| "Telenow does not know which voice and model to call with yet" | Create one agent in the organization first; the assistant's calls start from the same voice and model setup. |
| The assistant keeps asking you to sign in | The connection was disconnected, or the person who approved it left the organization or is no longer an owner or admin. Connect it again. |
| "This request was answered already" on the connect page | Allow was pressed twice (or the page was sent again); the first answer stands. Go back to the assistant to finish connecting. |
| The connect page shows a problem and a Return to … button | The assistant's request was wrong in a way it must be told about. Assistants that register themselves are not sent back automatically: check the address on the button is your assistant's before you press it. |
| "Telenow could not check this token just now" | A brief problem on Telenow's side (a 503). Try again in a moment; nothing needs to be connected again. |
| The connect page says the AI assistant calls agent is turned off, and Allow stays disabled while Place calls is ticked | The built-in agent that carries out one-off calls was turned off. Turn it on in the agent builder, or untick Place calls. |
| ChatGPT has no option to add a connector | Custom connectors need developer mode, on a paid plan; on a Business or Enterprise workspace an admin allows them first (Connect your assistant). |
| The assistant does not use Telenow | Turn Telenow on in the chat's tools menu, or name it: "Use Telenow to…". |
Calls
| What you see | What to do |
|---|---|
| "…is on this organization's Do-Not-Call list and cannot be dialed" | The number opted out and is never called. Remove it from the list in Telenow only if the person asked to be called again (Do-Not-Call). |
| "…has no recorded consent on file…" | Your organization requires proof of consent before dialing, and this number has none (Telephony compliance). |
| "Calls to this number may only be made between …" | The destination's legal calling window is closed — Telenow enforces the windows it knows (India's 09:00–21:00 IST today). The message says when it opens; ask the assistant to try then. |
"to must be a real international number…" | Give the number with + and the country code. Numbers that cannot exist in that country's numbering plan — a US number with area code 555, an Indian number that is too short — are refused. |
| "This idempotency_key was already used for a different call" | Each call needs its own key; a key is for retrying the same request. Use a new key. |
| "An earlier request with this idempotency_key has not finished placing its call…" | The first request is still dialing, or was interrupted. Check list_calls in a minute; if the call is not there, send the request again with a new key. |
| "The “AI assistant calls” agent is turned off in Telenow" | The built-in agent was turned off, so one-off calls described as a task stop. Turn it on in the agent builder, or call with one of your own agents (agent_id). |
| "This assistant may place at most N calls in 24 hours" | The connection's daily limit is reached. To raise it, connect the assistant again with a higher limit and disconnect the old connection. |
| The call ends after a few minutes | The connection's minutes-per-call limit, or the agent's own maximum length. |
get_call shows dialing for a long time | Nobody has answered yet. The call ends on its own if no one does, with outcome no-answer or busy. |
| No summary | summary_status: pending — the analysis is queued or running, usually a minute or two after the call ends; ask again. ready with no summary — the agent's analysis does not write one; its fields are in analysis. not_available — there will be none: the agent's post-call analysis is off, the call was too short or nothing was said, or the analysis failed or was abandoned. |
| "This agent needs values for: …" | The agent's prompt has a required {placeholder} the assistant did not pass. Give the value, or make the variable optional with a default. |
Agents
| What you see | What to do |
|---|---|
| "This agent was not created by this assistant" | An assistant changes only the agents it created. Change this one in the agent builder. |
list_voices: "Telenow has no voice list for …" | That provider's voices are not in Telenow's catalogue yet. Try another provider, or pick the voice in the agent builder. |
estimate_cost declines a realtime agent | Speech-to-speech agents are priced in the agent builder, which shows their cost per minute. |
| "…which reaches outside the call" — "…contains a URL (…)", "…is customllm, a custom endpoint", "…lets the agent book its own call-backs" and the like | The option needs Give agents tools: links (any text with ://, in the prompt too), custom model and speech endpoints, tools and tool steps, hand-offs to another agent (flow agent steps), transfers, publishing, messages after the call, call-backs. Connect the assistant again with it ticked, or add the option yourself in Telenow. |
"… is a credential (an API key or a secret)" | Keys and secrets are never accepted from an assistant, in any field. Add them in Telenow. A masked key (••••1234) read with get_agent may be sent back unchanged — the stored key stays. |
"… is a masked key copied from another agent" | A new agent starts without keys. Leave the field out; the owner adds the key in Telenow. |
"… holds a masked key this agent does not hold there…" | The step, tool or provider the key belongs to now points somewhere new, so the key does not follow it. Keep the old destination, or add a key for the new one in Telenow. |
"flow would be discarded at call time…" | The graph breaks a rule calls depend on — the message names it (a missing step, an exit without a condition, a transfer without a number). Fix it and send it again; nothing was saved. |
"flow has one step and no exits…" | Calls would run it as a single-context agent and ignore the step. Use create_agent, or add steps and the exits between them. |
warnings after creating a flow agent | The graph runs, but something in it may misbehave on a call — for example an exit that reads a variable no step collects. Read them to the user and fix what matters. |
Campaigns
| What you see | What to do |
|---|---|
| "This campaign was not created by this assistant" | An assistant starts and adds people only to campaigns it created. Start yours from Telenow, or let the assistant create a new one. |
| "This campaign is completed and cannot be started" | A completed or cancelled campaign cannot be started. Adding people to a completed one (add_campaign_targets) starts it running again to call them; a cancelled one takes nobody — create a new one. |
| "This campaign was cancelled, so nobody added to it would be called" | Create a new campaign for these people. |
| "The agent … does not exist in this organization or was deleted" | The agent_id given to create_campaign is wrong, or the campaign's agent is gone, so nobody added to it could be called. Use an agent from list_agents (in a new campaign, when it is the campaign's agent that is gone). |
| "This assistant may cause at most N calls in 24 hours (direct calls and campaign targets together)…" | The list would take the connection past its daily limit. Every new number in the list is counted when checking (Do-Not-Call numbers too); only the people actually added are kept against the limit. Send fewer, or connect again with a higher limit. |
| "At most 1000 targets per call" | Send the rest with add_campaign_targets, a page at a time. |
| "…not real international numbers…" | Every number needs +, the country code, and a number that country uses. The refusal lists examples. |
| "The agent requires values these targets lack…" | Give each person the values in their variables, or one value for everyone in variables. |
| "Telenow's hourly limit for this was reached" | Too many campaigns or rows in an hour. The message says when to retry. |
| Results stay empty | Nobody has finished a call yet — check the calling window and that the campaign is running. |
A result has analysis_status: pending and no fields | Its analysis is still running. Read that call with get_call in a few minutes (while the campaign's number is still one of the assistant's numbers); the next pages do not repeat it. |
added is lower than the list | duplicates: people already in the campaign — by phone number, whatever their result, so nobody is called twice — or repeated in the list (by number, or by an id you gave). do_not_call: numbers on your Do-Not-Call list. Where your organization requires consent, people without it are added but not dialed. |
| Campaign calls end after a few minutes | The minutes-per-call limit of the assistant's latest connection applies to the campaigns it created too — reconnecting with another limit changes it for them — as well as the agent's own maximum length. |
Still stuck
Disconnect under Workplace → Connected assistants, then connect again from your assistant — it costs nothing. The agents and campaigns it made stay its own when it comes back as the same client: an assistant that names itself with a client metadata URL (ChatGPT and Claude publish one) keeps them; one that registers afresh on each connection starts over, and what it made stays in Telenow for you. For anything else, contact support with the time of the request and the call or campaign id.