MCP server for Agentline — give your AI agent a phone number, email, SMS, and voice calls.
Agentline MCP demonstrates solid tool design with clear naming (verb_noun pattern), comprehensive descriptions (avg 180 chars), and well-structured input schemas. All 14 tools have explicit descriptions and typed parameters. However, output schemas are not formally documented, responses are inferred from code (asdict conversions, dict literals). Error handling returns structured dicts with error/status_code but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications in metadata. Parameter descriptions are good but some lack format constraints (e.g., E.164 format stated in description but not as regex pattern). Overall: production-ready definitions with room for schema formalization and error messaging.
Capture a code for an ALREADY provisioned number after submitting signup. First provision_number, record a UTC ISO timestamp, submit the phone in the signup form, then call this tool with that timestamp as since. Release the number separately when finished. This tool never provisions a number.
Provision a new email address for sending and receiving mail. Use this when signing up for services that require email verification — the agent gets a real inbox. `local_part` is the part before the @ (e.g. "my-agent"); leave null to auto-generate. Returns `id`, `email_address` (full address to paste into forms), `provider`, `status`.
Get status, transcript, and summary for a call. Call this after `make_call` to check whether the call has completed. Terminal statuses are `completed`, `failed`, `no_answer`, `busy` — anything else means the call is still in progress and you should poll again.
End an in-progress call immediately. No-op / error if the call already terminated.
List all email addresses currently provisioned on this account.
Output schemas not formally documented. Tools return dicts (asdict conversions, dict literals) but no JSON Schema definitions provided for responses. LLMs cannot reliably plan downstream tool calls or extract required fields.
Tool annotations missing. Risk classifications (WRITE, DESTRUCTIVE, READ_ONLY, REVERSIBLE) exist in metadata but are not exposed as readOnlyHint/destructiveHint/idempotentHint in tool definitions. Agents cannot infer safety properties.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
List all phone numbers currently provisioned on this account. Useful when the agent needs to find a number it forgot about, or audit what's active.
Place an outbound AI voice call. An AI agent answers when the recipient picks up and follows `prompt` as its system instructions; `first_message` is what it says on connect. Returns immediately with the `id` (call_id) — the call runs in the background. Poll `get_call(call_id)` to see status, transcript, and summary once it completes. `voice` is a Deepgram TTS voice (aura-asteria-en, aura-orion-en, …).
Provision a phone number the agent can use — for receiving SMS (e.g. 2FA), sending SMS, or placing/receiving voice calls. Use this BEFORE signing up for a service that asks for a phone number, or before making outbound calls. If `phone_number` is given, provisions that specific number; otherwise auto-searches by `area_code`. Returns a dict with `id`, `phone_number` (E.164 format — pass this to forms), `provider`, and `status`. Save the returned `phone_number` — you need it to later `release_number` or `wait_for_sms`.
Release a provisioned email address. Pass the `id` returned from `create_email_address` (NOT the full email string).
Release a previously provisioned phone number. Call this when done to avoid ongoing Telnyx monthly charges. Pass the E.164 `phone_number` returned by `provision_number`.
Send an outbound email from a provisioned address. `from_email` must be one of your provisioned addresses.
Send an outbound SMS from a provisioned number. `from_number` must be a number you already provisioned; `to_number` is the recipient in E.164 format (e.g. "+15551234567"). Outbound US SMS, international SMS, and inbound SMS all work — Agentline's 10DLC campaign is approved across all major US carriers.
Long-poll (blocking up to `timeout` seconds, max 180) for the next inbound email on a provisioned address. Use this right AFTER submitting a form that triggers an email — call this to catch the reply. `match` is an optional regex — only messages matching the pattern satisfy the wait. Any inbound message satisfies it when `match` is None. Returns the email dict (with `from`, `subject`, `body`, and auto-extracted `extracted_code` when present), or `{"message": null, "status": "timeout"}` if no email arrived in time.
Long-poll (blocking up to `timeout` seconds, max 180) for the next inbound SMS on a provisioned number. Use this right AFTER submitting a form that triggers an SMS — call this to catch the reply. `match` is an optional regex (e.g. `\d{6}` for a 6-digit code) — only messages matching the pattern satisfy the wait. Any inbound message satisfies it when `match` is None. Returns the message dict (with `body` and auto-extracted `extracted_code` when present), or `{"message": null, "status": "timeout"}` if no message arrived in time.
Error handling lacks recovery guidance. Errors return {error: str, status_code: int} but do not suggest next steps. E.g., 'Invalid phone number' should suggest 'Try provision_number() first' or 'Check E.164 format (+1XXXXXXXXXX)'.
Parameter format constraints stated in descriptions but not enforced as JSON Schema patterns. E.164 format, regex patterns (match param), and timeout bounds (max 180) are documented but not machine-parseable.
list_numbers and list_email_addresses lack pagination. No limit, offset, or cursor parameters. Large result sets could exhaust context window.