AI and MCP Services Addendum
This Addendum forms part of the Terms of Service and applies when you use any AI feature of the Services or connect an AI agent to the Services. If this Addendum conflicts with the Terms on an AI matter, this Addendum prevails.
1. What this Addendum covers
1.1 AI Features means features that use machine learning or large language models to process Customer Content, including call transcription, call summaries, sentiment and topic analysis, AI voice agents, and AI-assisted call flows.
1.2 MCP Server means our Model Context Protocol server at mcp.sipio.dev/mcp, which lets AI agents and assistants that you connect (“AI Agents”) call the Services’ tools, such as placing calls, sending messages, searching and buying Numbers, reading CDRs and changing configuration, using your API keys or OAuth tokens.
1.3 Output means transcripts, summaries, generated text or speech, and any other content produced by an AI Feature or AI Agent.
2. Your responsibility for AI Agent actions
2.1 An AI Agent connected to the MCP Server or the API acts on your behalf. You are responsible for every action an AI Agent takes with your API keys, OAuth tokens or other Credentials, exactly as if you had taken the action yourself. This includes:
- calls placed and messages sent, including to whom and with what content;
- Numbers searched, bought, released or ported;
- configuration, routing, call flow and trunk changes;
- data read, exported or deleted; and
- money spent from your Wallet, including through top-ups and auto-recharge.
2.2 This applies whether the AI Agent is provided by you, by sip.io, or by a third party, and whether or not the action was what you intended, including where the AI Agent was manipulated by content it processed (for example, prompt injection in a message, web page or document).
2.3 All rules in the Acceptable Use Policy, Voice Services Policy and Messaging Policy apply to actions taken by AI Agents, including consent, do-not-call and calling-hour rules.
3. Minimum scopes and human oversight
3.1 You must grant AI Agents the narrowest access they need. In particular, you should:
- create separate API keys or OAuth grants for each AI Agent, limited to the scopes it needs (for example read-only access to CDRs);
- avoid granting billing, Number purchase, sub-account or Credential-management scopes unless essential;
- set spending, channel, CPS and destination limits that cap what an AI Agent can do; and
- revoke access immediately when no longer needed or if you suspect misuse.
3.2 You must keep meaningful human oversight of AI Agents, including review and approval for consequential or irreversible actions such as bulk outbound campaigns, Number purchases or releases, large spend and data deletion. Where an MCP client offers confirmation prompts, keep them enabled for such actions.
3.3 You must not use AI Agents to make decisions with legal or similarly significant effects on individuals without the human review and safeguards required by applicable law.
4. Outputs may be inaccurate
4.1 AI Features and AI Agents are probabilistic. Outputs may be incomplete, inaccurate, biased or inappropriate, may misidentify speakers, and may differ for the same input.
4.2 You are responsible for reviewing Outputs before relying on them, and for decisions you make based on them. Outputs are not legal, medical, financial or other professional advice.
4.3 To the maximum extent permitted by law, we make no warranty as to the accuracy, completeness or fitness of Outputs, and we are not liable for actions taken by AI Agents or for decisions based on Outputs.
5. Use of Customer Content and model training
5.1 We use Customer Content processed by AI Features only to provide those features to you, to secure and support the Services, and to comply with law.
5.2 Our commitment: we do not use your Customer Content to train AI models, and we do not permit our third-party AI providers to use your Customer Content to train or improve their models. Our AI providers process Customer Content under their business or enterprise data terms, which include this restriction and limits on any retention for abuse monitoring.
5.3 Our AI providers, currently including xAI (Grok), are listed on the Subprocessors page, together with processing locations. AI processing of personal data is covered by the Data Processing Addendum.
5.4 As between you and us, you own the Outputs to the extent permitted by law, subject to the rights of any third party in the underlying content.
6. AI voice and disclosure laws
6.1 If you use AI-generated, synthetic or cloned voices on calls, you must comply with laws that apply to them. In the United States, the FCC has ruled that AI-generated voices are “artificial” voices under the TCPA, so calls using them require the prior express consent (and for telemarketing, prior express written consent) that the TCPA requires for artificial or pre-recorded voice calls, and must include the required identification and opt-out mechanisms.
6.2 You must disclose to call recipients that they are interacting with an AI system where the law requires it, including under the EU AI Act transparency obligations and US state laws on AI disclosure and bot interactions, and in any case where the recipient would reasonably expect to be told. Our recommended practice is to disclose at the start of every AI-handled call.
6.3 You must not clone or synthesise the voice of a real person without their documented, informed consent, and you must comply with laws on voice and likeness rights.
6.4 Recording and transcription performed by AI Features must comply with the Call Recording and Transcription Policy.
7. Prohibited AI uses
You must not use AI Features, the MCP Server or AI Agents to:
- impersonate any real person or organisation, or mislead recipients into believing they are speaking with a human or a particular person;
- create deepfakes of candidates, officials or private individuals, or carry out election-related deception or voter suppression;
- conduct fraud, scams, phishing, vishing, extortion or social engineering;
- make automated calls or send automated messages without the consent required by law;
- carry out large-scale scraping, enumeration of numbers, or attacks on systems;
- infer special-category data (such as health, sexual orientation or political opinions) about individuals from calls or messages without a lawful basis, or perform emotion recognition or biometric categorisation where prohibited, including in the workplace under the EU AI Act;
- engage in any practice prohibited by the EU AI Act or similar law; or
- circumvent rate limits, spending limits, safety filters or account isolation.
8. Availability and changes
8.1 AI Features and the MCP Server may depend on third-party providers and may change, be rate-limited, or be withdrawn. Unless stated otherwise, they are not covered by the Service Level Agreement, and features labelled beta or preview are subject to the beta terms in the Terms.
8.2 We may change the AI providers or models used for an AI Feature. We will update the Subprocessors page and give notice of new subprocessors as described in the Data Processing Addendum.
9. Security of the MCP Server
9.1 Connections to the MCP Server are authenticated with your API keys or OAuth tokens and are subject to the same account isolation, scopes and rate limits as the API.
9.2 You must protect the Credentials used by AI Agents as described in the Terms of Service and the Fraud and Toll-Fraud Policy. Treat any AI Agent that processes untrusted content (such as inbound messages or emails) as able to be manipulated, and limit its scopes accordingly.
10. Indemnity
You will defend, indemnify and hold harmless sip.io against claims arising from actions taken by your AI Agents, your use of Outputs, AI voice calls made through your Account, and your breach of this Addendum.