cloudtalk_add_call_tags
ChatGPTAttach one or more CALL tags to a call. Idempotent — tags already on the call are ignored, so re-running is safe. Call tags are a separate catalog from cloudtalk_list_contact_tags, whose ids will be rejected here; use cloudtalk_list_call_tags for the assignable catalog, or read ids from cloudtalk_get_call (call_tags[].id). The call is addressed by the numeric id from cloudtalk_search_calls (preferred — a later cloudtalk_get_call then reflects this change); a call uuid from cloudtalk_get_call is also accepted. Returns the call’s full tag set after the change. To see a call’s current tags without changing anything, use cloudtalk_get_call.
cloudtalk_add_contact_note
ChatGPTAdd a note to a contact. Notes are additive; existing notes are never touched. Returns the created note.
cloudtalk_assign_tool_to_voice_agent
ChatGPTAssign a Voice Agent tool to a Voice Agent so the agent can call it mid-conversation. Get the tool id from cloudtalk_list_voice_agent_tools and the agent id from cloudtalk_list_voice_agents. Assigning a tool that is already assigned is a safe no-op. It takes effect immediately: use it only when the user asked for it.
cloudtalk_bulk_tag_contacts
ChatGPTAssign tags to many contacts in one synchronous call, up to 500 contacts per call; more is rejected, so split larger sets into batches yourself. mode "append" (the default) adds the tags alongside existing ones; mode "replace" overwrites each contact's tag set with exactly these tags. Returns requested, ok, and an errors entry (contact_id, code, message) for each contact that failed; it is not all-or-nothing. This changes many contacts at once, and replace mode also removes their other tags: use it only when the user explicitly asked for it.
cloudtalk_create_campaign
ChatGPTCreate a dialer campaign. name and mode are required. A policy bundle you omit is stored empty rather than filled in, so cloudtalk_get_campaign returns nothing for it and the campaign runs on the engine's built-in behaviour for those settings. The campaign starts in initializing and dials nothing until it is set to active: attach target tags with cloudtalk_replace_campaign_associations, then start it with cloudtalk_set_campaign_status, or create, attach and start in one call with cloudtalk_launch_campaign_from_source. Calling it again creates a duplicate campaign, so call it once. It takes effect immediately: use it only when the user asked for it.
cloudtalk_create_contact
ChatGPTCreate a contact, or update the existing one that the company's contact merge setting matches. The setting is phone number, email, phone number and/or email (the default), or do not merge. When it matches an existing contact, that contact is updated instead of duplicated: non-empty fields sent replace the stored ones (except source, which keeps its stored value), custom_fields replaces the whole stored set, phone numbers, emails and tags are added, and a matched lead becomes a contact. Under do not merge, or when the match is on a key the setting does not use, a new contact is created; to avoid a duplicate there, look the contact up with cloudtalk_list_contacts first. name is required; all other fields are optional. Returns merged (true when an existing contact was updated, false when a new one was created; absent when the contacts service did not say) and the contact in the same envelope as cloudtalk_get_contact, including contact.custom_fields — read it back to confirm which custom field values were stored.
cloudtalk_create_knowledge_base
ChatGPTCreate a Voice Agent knowledge base. type "custom" stores text directly and is active right away; type "url" ingests a public web page asynchronously, and the knowledge base starts in the SYNCHRONIZING status. Returns the created knowledge base. To link it to an agent, call cloudtalk_update_voice_agent with the agent's full knowledgeBaseIds list plus this id, since the list is replaced as a whole. It takes effect immediately: use it only when the user asked for it.
cloudtalk_create_voice_agent
ChatGPTCreate a new Voice Agent configuration; it decides what the agent says to the people it talks to. To change an existing agent, use cloudtalk_update_voice_agent instead. For required values the user did not give (such as voice, temperature, stability and similarity), read an existing agent with cloudtalk_get_voice_agent rather than guessing. The server validates the configuration; if it is invalid the tool returns the validator's specific errors and nothing is saved. It takes effect immediately: use it only when the user asked for it.
cloudtalk_create_voice_agent_tool
ChatGPTCreate a Voice Agent tool: an HTTP action a Voice Agent can call mid-conversation. Provide name, method, url and awaiting_response_message; optionally headers, query/body params, and voice_agent_config_ids to assign it. Header and parameter values may hold credentials for the user's endpoint; they are sent to CloudTalk but never returned by any tool. If the server rejects the tool it returns the validator's specific errors and nothing is saved. It takes effect immediately: use it only when the user asked for it.
cloudtalk_delete_campaign
ChatGPTDelete a campaign. This cannot be undone: its status becomes deleted and the campaign can never be reactivated. This is not archiving: to archive, park, pause or retire a campaign, use cloudtalk_set_campaign_status with archived (reversible, exits back to inactive) or inactive instead. It deletes immediately: use it only when the user explicitly asked to delete this campaign.
cloudtalk_get_agent_performance
ChatGPTAgent 360°: aggregate one agent's activity over a time window — counts by direction, answered/missed, talk-time stats, top contacts, top tags. The aggregates cover only the calls fetched, which max_calls caps: aggregates.calls_analyzed is that sample size, and window.truncated says when there are more. window.calls_total_in_window reports how many calls the agent really has in the window, so never present calls_analyzed as that total. A null calls_total_in_window means that count could not be read; it never means zero. The same rule applies to talk time: a null total_talk_seconds or avg_talk_seconds means no answered call in the window reported a duration, so the figure is unknown rather than zero — do not present it as an idle agent. A zero means either that no call was answered in the window - check aggregates.answered - or that answered calls genuinely lasted no time. Compare aggregates.talk_reported_count with aggregates.answered to see how much of the window the figures actually cover; top_contacts[].talk_seconds follows the same rule and is null when no call with that contact reported a duration. When only some answered calls reported a duration, total_talk_seconds sums those and is a lower bound. Voicemail counts are not available and report as null. Optionally samples recent calls for AI sentiment distribution. Resolve the agent by user_id, email, or extension.
cloudtalk_get_call
ChatGPTFetch full detail for a single call: contact, agent, tags (call_tags, each {id, name}), rating, call timings, call flow steps (IVR, queues, transfers), topics, sentiment, and call score. Does not include summary or smart notes — use cloudtalk_get_call_insights for those.
cloudtalk_get_call_insights
ChatGPTBundle the Conversation Intelligence signals for one call: summary, overall sentiment, topics, smart notes (action items, risks/issues, customer feedback), and talk-listen ratio. Slower than fetching a single signal — use include to request only what you need. Results are cached. Set include=["summary","sentiment"] etc. to fetch a subset.
cloudtalk_get_call_recording
ChatGPTGet a playable link to a call's audio recording. Returns a signed URL valid for UP TO 12 hours, sometimes much less — fetch it when the user actually wants to listen, not speculatively, and re-fetch rather than storing it. A call with no recording returns url: null with reason "no_recording", which is a normal answer rather than an error. Use "download" instead of the default "play" only when the user explicitly wants to save the file; both are audited. For the words spoken use cloudtalk_get_call_transcript, and for summary/sentiment cloudtalk_get_call_insights — those are cheaper and usually what is wanted.
cloudtalk_get_call_transcript
ChatGPTFetch the verbatim transcript of a call as time-coded segments (start/end seconds, speaker, text), plus the speaker map and detected language. Use this when you need the actual words spoken; prefer cloudtalk_get_call_insights for summary/sentiment/topics. Paginate long calls with limit/offset. A call with no transcript returns transcription: null with reason "no_transcript". That is a normal answer, not an error, and not a signal to retry: it covers calls that were never analysed, including every call shorter than 30 seconds. Calls handled by a Voice Agent are transcribed like any other call once they last 30 seconds or more.
cloudtalk_get_campaign
ChatGPTGet one dialer campaign by id: name, mode, status, contact-source order, the policy bundles, and linked call script / survey. Returns null when the campaign does not exist or belongs to another company. Note: CRM connections are not returned — this read does not expose them.
cloudtalk_get_campaign_associations
ChatGPTA campaign's contact sources and assignees: target tags (with priority, label and whether a CRM connection owns the row), agent groups, and agents. Read this before cloudtalk_replace_campaign_associations — that write replaces a whole kind, so you need the current set first.
cloudtalk_get_campaign_statistics
ChatGPTGet aggregate dialing statistics for one dialer campaign: the contact-state funnel (pending, scheduled, completed, …), dialed / answered / answer_rate, conversions, the human / voicemail / other-machine split, the disposition breakdown and call durations. Without a window the figures are all-time; with date_from and date_to (both required together, ISO 8601, last 13 months only) they cover that half-open window, and range echoes it. A date-only date_to includes that whole day. A window always requests durations; for all-time they are opt-in via include_durations. call_durations is null when the dialer could not read them, which means unknown, not zero. A windowed read is expensive, so do not fan it across many campaigns. Returns statistics null when the campaign does not exist or belongs to another company. Use cloudtalk_list_campaigns to find campaign ids.
cloudtalk_get_contact
ChatGPTFetch full detail for a single contact: profile, address (incl. country name + code), all phone numbers and emails, tags, custom attributes (custom_fields, keyed by field name; an empty object means the contact has none), external CRM URLs, and favourite agent. Use cloudtalk_list_contacts to find a contact_id first.
cloudtalk_get_contact_interactions_summary
ChatGPTCompact, token-lean digest of one contact's interaction history — call mix, last contact, top tags, plus a small AI sample over the most recent answered calls (per-call summary + sentiment, a top-topics roll-up, and a sentiment trend). Every figure under interactions is computed over the calls actually fetched, which max_calls caps: interactions.calls_analyzed is that sample size and interactions.truncated is true when the contact has more. Read window.calls_total_in_window for how many calls the contact really has in the window, and never present calls_analyzed as a total. A null calls_total_in_window means that count could not be read; it never means zero. total_talk_seconds sums each answered call's talking time, so ring and waiting time are excluded; it is null when answered calls exist but none reported a duration, which means unknown rather than zero, and a lower bound when only some reported. Use this for a quick 'what's the story with this contact?'; use cloudtalk_get_contact_journey for the full 360° timeline. Resolve a contact by id or phone. When contact is null the response always carries a reason, and the interaction figures may still be present: reason 'no_contact_with_phone' means no contact holds that number, while 'contact_not_found' means the contact could not be read — which for a contact_id from another company leaves the figures at zero, so treat them as unknown rather than as that contact's real activity. This differs from cloudtalk_get_contact_journey, which returns nothing but the reason.
cloudtalk_get_contact_journey
ChatGPTCustomer 360°: aggregate everything we know about one contact — profile, the calls in the window with summary aggregates (counts, talk time, top tags), agent notes, and CRM activities. aggregates.total_talk_seconds sums each answered call's talking time and is null when answered calls exist but none reported a duration — unknown, not zero, and a lower bound when only some reported. aggregates.avg_talk_seconds averages over the calls that reported, so it is null in that same case, and aggregates.talk_reported_count against aggregates.answered says how partial both figures are; whereas a timeline row's total_time is that call's composite duration — ring plus talking plus wrap-up inbound, talking plus wrap-up outbound — so summing the timeline gives a larger number and neither is wrong. The aggregates cover only the calls fetched, which max_calls caps: aggregates.calls_analyzed is that sample size, and window.truncated says when there are more. window.calls_total_in_window reports how many exist, and a null there means that count could not be read — it never means zero. Optional sentiment sample over recent calls. Resolve a contact by id or by phone number; this is an expensive call, so cap it with a date range.
cloudtalk_get_voice_agent
ChatGPTGet the full configuration of a single Voice Agent by id (the voice_agent_id from cloudtalk_list_voice_agents). Returns the complete config document. Its optimizedSettings lists the values the platform would pick for this agent's current setup (provider, llm, ElevenLabs turn settings, the latest minorVersion); they are never applied on their own. To change the agent, pass only the changed fields to cloudtalk_update_voice_agent; to adopt optimizedSettings, pass "apply_optimized_settings": true there instead of copying the values.
cloudtalk_health_check
ChatGPTConfirm the CloudTalk MCP is reachable and report which company and user your credentials resolved to. Makes no API calls of its own. Run this first when a tool is failing, to tell an auth or tenant problem from a data problem.
cloudtalk_initiate_call
ChatGPTStart an outbound click-to-call: CloudTalk first rings the chosen agent, and once the agent picks up (within ~20s) it automatically dials the customer number. The agent must be online and not already on a call (use cloudtalk_list_agents to find an agent_id). To have a Voice Agent make the call instead, use cloudtalk_trigger_aiva_call. Success means the agent is being rung, not that anyone answered, and calling it again places another call. This is an irreversible, outward-facing action that rings the agent immediately: use it only when the user explicitly asked for this call.
cloudtalk_launch_campaign_from_source
ChatGPTCreate a campaign, attach its contact sources, and start it, all in one call. Pass at least one target_tag: only target tags decide who is dialed (queue groups and agents decide who dials), so a launch without them goes active and calls nobody. Returns the campaign id and which of the three steps completed. It is not atomic: if a later step fails the campaign still exists, left in initializing so it dials nothing, and the response says how far it got. Pass activate=false to stage it without dialing. Calling it again starts a duplicate campaign, so call it once. Activating immediately places real outbound calls: use it only when the user explicitly asked to start this campaign.
cloudtalk_list_agents
ChatGPTList agents in the account: id, name, email, extension, availability status, outbound-calling status, and default phone number. Use this to translate a user_id from calls into a person, look up an extension, or find which agents are online.
cloudtalk_list_call_tags
ChatGPTList the CALL tags this company can attach to a call — every id returned is one cloudtalk_add_call_tags accepts. Use it to find a tag_id for the call-tag write tools or for cloudtalk_search_calls. Distinct from cloudtalk_list_contact_tags, the separate contact/CRM catalog whose ids do not resolve for calls. Cursor-paginated: pass the returned next_cursor back to walk the catalog.
cloudtalk_list_campaigns
ChatGPTList outbound dialer campaigns (id, name, mode, status, contact-source order, policy bundles, linked call script / survey). Cursor-paginated — pass the returned next_cursor to page; optionally filter by status or mode. Use cloudtalk_get_campaign for one campaign plus its associations.
cloudtalk_list_contact_tags
ChatGPTList or SEARCH the company CONTACT tags (the tag catalog used for contacts and contact segments), to pick tag names for cloudtalk_bulk_tag_contacts or to show which segments exist. Pass query to search by name substring — do that rather than paging: the catalog runs to tens of thousands of mostly machine-generated names (22,071 on one test company), so paging it to find a segment is not practical, while a substring search returns a handful. tag_type narrows to contact or integration and applies only alongside query, because the listing cannot filter by type; do not count on it to hide bulk-generated tags, which on the company measured are all typed contact. Searching is cursor-paginated and reports no total; listing is page-paginated with counts. NOT a source of ids for calls: call tagging and cloudtalk_search_calls use a separate call-tag catalog, so an id from here will not resolve there.
cloudtalk_list_contacts
ChatGPTList or search contacts, in one of three lanes. email is an EXACT address lookup and is the one to use before creating a contact — it answers "does this address already exist" and returns every contact holding it. A keyword query (2..200 chars) is a TEXT search across name, number and email, so it also matches records that merely mention the text and is the wrong tool for a deduplication check. Omit both to list contacts (optionally narrowed by tags or country). Tag filtering differs per lane: listing matches ALL the tags given by default (switchable with tag_match), keyword search always matches ANY of them, and email takes no filters at all. Returns a paginated list with id, name, company, title, industry, the first number, and one email — the contact's first, except on the email lane where it is the address that matched. Use cloudtalk_get_contact for every number and address a contact holds.
cloudtalk_list_groups
ChatGPTList agent groups (queues/teams). Useful when you need a group_id for filtering or to map realtime stats from cloudtalk_realtime_group_stats back to a team name.
cloudtalk_list_knowledge_bases
ChatGPTList the account’s Voice Agent knowledge bases: id, name, type (CUSTOM/URL/FILE), status, and which agent configs they are attached to. Use an id in a Voice Agent’s knowledgeBaseIds. Paged with limit/page; pagination.items_count is the unpaged total and pagination.next_page is precomputed or null. limit trims the response after fetching, so it bounds context size rather than cost.
cloudtalk_list_numbers
ChatGPTList phone numbers (DIDs) attached to the account: id, country/area code, friendly name, E.164 caller ID, and which channels each number supports (voice, SMS, MMS, WhatsApp). Useful to translate the number a call came in on into a friendly name, or to find a company number to send an SMS from — for that, pick one with "sms_supported": true, because cloudtalk_send_sms is rejected by any other sender.
cloudtalk_list_sms_conversations
ChatGPTList SMS conversations (threads) for the acting user: contact number, unread count, and a preview of the last message. Filter by a specific customer number or unread-only. Results are scoped to that user’s SMS visibility. Use cloudtalk_list_sms_messages to read the full thread.
cloudtalk_list_sms_messages
ChatGPTList the messages within an SMS conversation (thread), newest-first, with direction and content. Scoped to the acting user’s SMS visibility. Get the conversation_id from cloudtalk_list_sms_conversations first.
cloudtalk_list_voice_agent_tools
ChatGPTList the account’s Voice Agent tools — HTTP actions a Voice Agent can call mid-conversation. Each tool includes voice_agent_config_ids (the Voice Agents it is assigned to). Optionally filter to the tools assigned to one Voice Agent. Use a tool’s id with cloudtalk_assign_tool_to_voice_agent / cloudtalk_unassign_tool_from_voice_agent. Paged with limit/page; pagination.items_count is the unpaged total (after any voice_agent_id filter) and pagination.next_page is precomputed or null. limit trims the response after fetching, so it bounds context size rather than cost.
cloudtalk_list_voice_agents
ChatGPTList the account’s Voice Agents (also known as AIVA): id, name, and direction (inbound/outbound). Use this to find the voice_agent_id required by cloudtalk_trigger_aiva_call, cloudtalk_get_voice_agent and cloudtalk_search_calls (to list one Voice Agent’s calls). Paged with limit/page; pagination.items_count is the unpaged total and pagination.next_page is precomputed or null. limit trims the response after fetching, so it bounds context size rather than cost.
cloudtalk_log_contact_activity
ChatGPTLog a CRM-style activity against a contact (e.g. an order, email, or custom external event). Activities are additive; existing activities are never touched. Returns the created activity id.
cloudtalk_realtime_group_stats
ChatGPTLive operational stats per group/queue: agents online, answered/unanswered counts, abandon rate, average and max waiting times, current waiting queue depth. Always fresh — not cached. Use for live monitoring questions like "how busy are we right now?".
cloudtalk_remove_call_tags
ChatGPTDetach one or more catalog tags from a call. Idempotent — tags not on the call are ignored, so re-running is safe. Only the tag ids you name are removed; the call’s other tags are untouched. Returns the call’s full tag set after the change. Reversible with cloudtalk_add_call_tags.
cloudtalk_replace_campaign_associations
ChatGPTSet a campaign's contact sources and assignees: target tags decide who is dialed, queue groups and agents decide who dials. A kind you pass (target_tag, queue_group, agent) replaces its current set in full: send the complete desired set, not additions, and an empty array clears the kind. A kind you omit is read from the campaign and kept unchanged; if that read fails, nothing is written. Read cloudtalk_get_campaign_associations first. The result names exactly what was written. Target tags synced by a CRM connection are server-owned: passing one here changes nothing (it is dropped from the write and reported back in crm_owned_ignored), and omitting one never removes it. On an active campaign, new target tags start real outbound calls immediately: use it only when the user explicitly asked for this change.
cloudtalk_search_calls
ChatGPTList calls with filters (date, contact, agent, group, Voice Agent, tag, status, direction, external number, country, without-tags, sentiment). Returns compact summaries (id, date, contact, external number, agent, Voice Agent, direction, duration, tags as {id, name}), newest first. Voice Agent calls are included: a call a Voice Agent handled has voice_agent set and agent null, and a person's call has agent set (plus voice_agent when a Voice Agent transferred the call to them). Filter by voice_agent_id for one Voice Agent's calls, or by user_id/group_id for people's. Whenever Voice Agent calls are included, items_count counts Voice Agent steps and adds people's calls, so it can exceed the calls listed: a call with more than one Voice Agent step, or one a Voice Agent handed to a person, counts more than once. Covers EXTERNAL calls only — internal (agent-to-agent) calls are not in this call log, so a total here can read lower than the same period in the CloudTalk Dashboard, which counts them; say so rather than presenting a count as the account total. Each row's total_time is a composite duration, not the connected time: ring plus talking plus wrap-up on an inbound call, talking plus wrap-up on an outbound one. The CloudTalk public API reports a billsec for the same call that counts the connected part only, so the two do not match — compare talking_time, the closest counterpart. Requires a date range unless you pass an exact call_id: omitted bounds default to roughly the last three months, and a wider span is rejected rather than shortened — move the window back (the span is capped, the start is not) for older calls. Pagination depth is also capped, so reach older or narrower results by moving the window or adding a contact/agent/group filter rather than paging deep. items_count and page_count are answered on the FIRST page only and are null on later ones: a later null means the count was omitted, not zero. The page-1 figure is a snapshot taken at that request and can drift if new calls match while you page, so report it as of page 1. Use cloudtalk_get_call for full detail and cloudtalk_get_call_insights for AI insights.
cloudtalk_send_sms
ChatGPTSend an outbound SMS to a customer from one of the company's phone numbers. The "sender" must be a company-owned number with SMS enabled (use cloudtalk_list_numbers and pick one whose "sms_supported" is true). On a send, "dashboard_url" links to the Messages report, not to this individual message, because SMS has no page of its own. Call it once: calling it again sends the message again. This is an irreversible, outward-facing action that sends immediately: use it only when the user explicitly asked to send this message.
cloudtalk_set_campaign_status
ChatGPTStart, pause, or archive a campaign by moving its status. active immediately starts placing real outbound calls to the contacts behind the campaign's target tags; a campaign with no target tags goes active and calls nobody, and agents or queue groups do not change that (they decide who dials, not who gets dialed), so attach target tags with cloudtalk_replace_campaign_associations first. inactive pauses it; archived shelves it (reversible, back to inactive only). Turning a campaign off never interrupts calls already in progress: running sessions drain on their own. Illegal transitions are rejected. To create a campaign, attach its target tags and start it in one call, use cloudtalk_launch_campaign_from_source instead. Set active only when the user explicitly asked to start this campaign.
cloudtalk_trigger_aiva_call
ChatGPTStart an outbound Voice Agent call to a customer: the Voice Agent dials the number and holds the conversation itself. Use cloudtalk_list_voice_agents to get a voice_agent_id first. To connect a human agent instead, use cloudtalk_initiate_call. It returns status "queued" and a session_id, not whether anyone answered, and calling it again places another call. This is an irreversible, outward-facing action that dials immediately: use it only when the user explicitly asked for this call.
cloudtalk_unassign_tool_from_voice_agent
ChatGPTUnassign (remove) a Voice Agent tool from a Voice Agent so the agent can no longer call it. It removes only this link: the tool itself stays, and other agents keep it. Unassigning a tool that is not assigned is a safe no-op. It takes effect immediately: use it only when the user asked for it.
cloudtalk_update_campaign
ChatGPTUpdate a campaign's configuration. Only the fields you pass change, but a policy bundle you pass is replaced in full, not merged key by key: any key you leave out of it is dropped, and the campaign then runs as though that setting had never been configured (cloudtalk_get_campaign stops returning it). Send the complete bundle from cloudtalk_get_campaign. Pass null for call_script_id or survey_id to clear that link. Use cloudtalk_set_campaign_status for status changes, and cloudtalk_replace_campaign_associations to change target tags, queue groups or agents. It takes effect immediately: use it only when the user asked for it.
cloudtalk_update_contact
ChatGPTUpdate an existing contact. Only the fields you pass are changed; omitted fields are left untouched. Array fields (tags, phone_numbers, emails) and custom_fields are replaced as a whole, not merged: passing one tag drops every other tag on the contact. To add tags without removing others, use cloudtalk_bulk_tag_contacts with mode "append"; one contact is fine. Returns the updated contact (re-fetched) in the same envelope as cloudtalk_get_contact, including contact.custom_fields; read it back to confirm which custom field values were stored. It takes effect immediately: use it only when the user asked for it.
cloudtalk_update_voice_agent
ChatGPTChange a Voice Agent's configuration. Pass voice_agent_id and only the fields you want to change: the tool reads the agent's current configuration and merges your fields onto it, so anything you leave out keeps its stored value. Collections are replaced, not merged: to change one entry of skills, scenarios, guardrails, secondaryLanguages or knowledgeBaseIds, read the current list with cloudtalk_get_voice_agent and pass the whole list back. To adopt the optimizedSettings that cloudtalk_get_voice_agent reports, pass "apply_optimized_settings": true; its description says when that must be a save of its own. The result lists the fields it changed; invalid configs are rejected with the validator's errors and nothing is saved. Not to be confused with cloudtalk_update_voice_agent_tool, which edits a Voice Agent tool (an HTTP action), not an agent. It takes effect immediately: use it only when the user asked for it.
cloudtalk_update_voice_agent_tool
ChatGPTUpdate an existing Voice Agent tool. Only the fields you pass are changed; anything you omit is preserved from the current tool. No tool returns a tool's headers or parameters, because their values may be credentials, so you cannot read them back: omit headers, query_params and body_params to keep them. A passed array replaces the whole set, so passing one header drops the others. To change only which Voice Agents a tool is assigned to, use cloudtalk_assign_tool_to_voice_agent or cloudtalk_unassign_tool_from_voice_agent instead. Invalid tools are rejected with the server's errors and nothing is saved. It takes effect immediately: use it only when the user asked for it.