Skip to main content
Glama

Dialyne

Place a phone call in France

place_phone_call
Destructive

Place a real outbound phone call to a number in France (+33) and get back a written result: a voice agent carries out the task you describe (book a table, move an appointment, ask a business a question, follow up on an order or a claim). Compliance: the service's rules require the voice agent to disclose it is an AI assistant at the start of every call (EU AI Act art. 50), and calls are placed only on the user's explicit request. The only exception is a callee the user confirms already knows an AI will call (ai_disclosure; not recommended).

The voice agent opens by saying it is an AI assistant calling on behalf of the user; if the callee objects to the call, it ends the call politely and the number is never called again by this service. It navigates phone menus (voice or keypad), waits on hold and talks with the person who answers. No audio is returned or recorded: you get a written outcome, a summary and, on request, the transcript (text kept 90 days at most).

When to use: the user explicitly asked you to phone a specific French business or person, and the number and the task are confirmed.
Number found by you: if the number did not come from the user (you found it by web search, in memory or in a document), STOP before calling: show the number to the user, ask "Shall I call this number?" and call only after they confirm in a later message.
Missing details: do not quiz the user for every detail before calling. Give the facts you already have in task; if the callee asks for something the user did not give, get_call shows it as pending_question and you ask the user then.
When not to use: marketing, sales prospecting, surveys, debt collection, mass or repeated calling (unlawful in France and refused); impersonating another person, an authority or a bank; threats or harassment; asking the callee for their codes, passwords, card details or a money transfer, gift cards or crypto (claiming money owed to the user is fine); dictating a payment code (3-D Secure, voucher) or a full card number with its security code; emergencies (emergency numbers are refused: tell the user to dial them directly). When you decline, tell the user why in one sentence (e.g. prospecting calls are unlawful in France and refused by this service).
Codes and payment details: the user decides what to give. A verification code, password or card number the user gives you may go in task or in answer_call_question; good practice is to share it only when the call needs it, mark it [secret: value], and prefer transfer_call_to_user for a payment.

Behavior: places a real call that cannot be taken back once answered. It is paid from the account's prepaid credit, per second of conversation; busy or unanswered calls are free and retried once after 15 minutes, within 08:00-21:00 callee's time (on lines where every call needs a new request from the user, get_call shows callback_suggested instead: ask the user before calling again). The call is placed now (or at scheduled_at) whatever the hour: the user decides. Returns immediately with a call_id while the call runs in the background (usually 1-10 minutes): then call get_call (wait_seconds=25) until done=true, answer any pending_question with answer_call_question within 45 seconds (hold_call_question asks the callee for one more minute), and report the outcome and summary to the user. To retry safely after an error or timeout, reuse the same idempotency_key.

Input Schema

NameRequiredDescriptionDefault
taskYesWhat the call must achieve, in plain English or French (10-2000 characters): the goal, on whose behalf (the user's name), the facts the callee may ask for (dates, booking or customer references) and the limits (what to accept or refuse, e.g. 'any time 19:30-21:00 is fine, no deposit'). The voice agent follows it, including giving a code or card detail the user chose to share; write such a value as [secret: value] so it is masked in stored transcripts and logs (card numbers are masked automatically). Refused: asking the callee for codes, passwords, card details or a money transfer (claiming money owed to the user is fine), and dictating a payment code or a full card number with its security code.
voiceNoOptional voice: '1' Solène (female, calm), '2' Gaspard (male, warm), '3' Coralie (female, lively), '4' Marius (male, deep), '5' Bella (female, premium, most expressive). Omit to use the account default.
languageNoLanguage spoken on the call: 'fr' (French), 'en' (English) or 'auto' (starts in French, switches to English if the callee speaks English). Omit to use the account default (usually French).
phone_numberYesNumber to call, French only. Copy it exactly as the user or the source wrote it: national format '01 42 00 00 00' is accepted and converted for you, so do not rewrite it yourself; E.164 '+33' followed by 9 digits (e.g. +33142000000) also works. Landlines (01-05, 09) and mobiles (06, 07) are allowed; emergency, short and premium-rate numbers are refused. A number you found yourself must be confirmed by the user before you call.
scheduled_atNoWhen to PLACE the call (not the time of the booking or appointment: that goes in task). Omit to call now, the normal case. Only for a call the user wants placed later, e.g. 'call the vet tomorrow at 10': ISO 8601 with time zone, e.g. 2026-11-03T14:00:00+01:00. The call is placed at that time, whatever the hour: the user decides.
ai_disclosureNo'standard' (default, recommended): the voice agent says it is an AI assistant in its opening line. 'callee_already_informed': leave that mention out of the opening line. Before using it, explain to the user that removing the AI disclosure is not recommended and is allowed ONLY when the person called already knows they will talk to an AI (e.g. a relative the user warned beforehand); use it only if the user then confirms it, and put their confirmation in callee_informed_confirmation. Never use it on your own initiative. Even then the voice agent never denies being an AI, tells the truth if asked, and announces it is an AI if someone else answers. Every use is logged.standard
outside_hoursNoObsolete, accepted and ignored: no calling-hours window is imposed any more. Omit it.
voice_settingsNoOptional tuning, only if the user asks for it (omit otherwise: the account settings, already optimal, apply). Basic: address_form 'vous' (default) or 'tu'; expressiveness 'natural' (default) or 'plain'; hesitations (default true, light 'alors…'); background 'none' (default), 'office', 'cafe', 'street', 'outdoor'; voicemail 'leave_message' (default) or 'hang_up'. Advanced (bounded, else rejected): response_delay 0.7-1.5 s step 0.1 (max wait before answering, default 1.0); silence_prompt_after 4-10 s (says 'Allô ?' after a silence, default 6); interruption_min_speech 0.3-1.0 s (default 0.5); model 'v4_turbo' (default) or 'flash_v2_5'; stability 0.30-0.60; similarity 0.50-0.90; human_touch 0.05-0.40 (hesitation density); background_volume 0.03-0.20; flash only: style 0-0.30, speed 0.90-1.10, speaker_boost. The AI disclosure is never altered.
idempotency_keyNoOptional unique key (letters, digits, '-', '_', '.', ':'; max 64) for this call request. Retrying with the same key within 24 h returns the original call instead of placing a second one.
callee_informed_confirmationNoRequired with ai_disclosure='callee_already_informed' (10-300 characters): the user's confirmation, in their words, of who the callee is and how they know an AI will call (e.g. 'My sister Lea, I told her this morning an AI would call her for me'). Kept as proof.

Output Schema

NameRequiredDescriptionDefault
statusNo
call_idNo
outcomeNo
summaryNo
languageNo
next_stepNo
phone_numberNo
scheduled_forNo
duplicate_requestNo

Schema Changelog

Changes observed during successful MCP inspections.

  1. First observed

TDQS

A4.2/5.0
Behavior3/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already flag destructiveHint=true, openWorldHint=true, readOnlyHint=false and idempotentHint=false, so the safety profile is partly given. The description still adds substantial non-schema behavior: an irreversible real call once answered, per-second prepaid billing, free retry of busy/unanswered calls after 15 minutes, a 08:00-21:00 window that does not block scheduled calls, background execution returning a call_id within 1-10 minutes, and the 45-second answer window. The apparent tension between destructiveHint/idempotentHint=false and the recommended reuse of idempotency_key is a per-request idempotency semantics point, not a contradiction.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

It is long, but front-loaded with the purpose and organized into clear labeled blocks (Compliance, When to use, When not to use, Codes and payment details, Behavior) that match the decision path of a high-stakes mutation tool. Some content is duplicated with the task parameter description (the refused-payment list appears twice), which is the main thing keeping it from a 5.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a destructive, open-world, paid, asynchronous tool with 10 parameters, the description covers authorization (explicit user request), compliance (EU AI Act art. 50), cost, irreversibility, the full async follow-up loop, and retry strategy. An output schema exists, so it does not need to explain return values, and it stays appropriately high-level there.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so baseline is 3 and the schema already carries voice personality labels, language modes, phone format rules and voice_settings ranges. The description adds cross-cutting semantics the schema cannot: the [secret: value] masking convention, why the AI disclosure default should not be changed, and the requirement to confirm a self-found number before calling.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The opening sentence gives a specific verb and resource ("Place a real outbound phone call to a number in France (+33)") plus the return form ("a written result"), and the parenthetical lists concrete tasks (book a table, move an appointment). It is clearly distinguishable from siblings like place_batch_calls and get_call, which the description also names later.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

There are explicit when-to-use criteria (user explicitly asked, number and task confirmed) and explicit when-not-to-use rules (prospecting, surveys, debt collection, impersonation, emergencies, refusing to reveal codes). It also routes the agent to siblings by name: get_call for polling/pending_question, answer_call_question within 45 seconds, hold_call_question, transfer_call_to_user for payments.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Try in Browser

Glama MCP Gateway

Add one secure layer between your agents and this server.