AI AGENT STORE / SERVICES

MEDIA PROCESSING & DELIVERY · FOR AI AGENTS

Tavus CVI

Create and operate conversational video agents through API, MCP and CLI.

Tavus Conversational Video Interface supplies a managed pipeline for video conversations with voice, visual interaction and tool calls. Its development MCP and CLI let coding agents configure PALs, connect application tools and start test conversations.

Your integration checklist

What you need from your owner
Create a Tavus developer account, configure credentials and billing, and choose a face, voice and permitted application tools. Set session duration, concurrency and spending limits.
Authentication
Development MCP uses OAuth or a bearer API key at https://mcp.tavus.io/mcp. REST uses x-api-key. Match credentials to the intended environment.
Your first task
Configure a bounded conversation with one read-only application tool and verify the tool result reaches the conversation before enabling customer workflows.
Billing & payment
Conversation usage is billed from session creation, including time before a participant joins. Minimum billing, rounding, concurrency limits and paid-plan overages apply; check the current plan before creating sessions.
Agent payment support
Not verified. Programmatic usage does not establish autonomous checkout support.
Operational limits
The documentation MCP only retrieves docs; it does not operate conversations. Development tools default to production. A created conversation is not proof that a participant joined or that an application action completed.

ORIGINAL AI AGENT STORE INTEGRATION PLAYBOOK

Bound the session lifecycle before connecting application actions

Editorial recommendation · Reviewed 2026-11-02

When to choose this service

Choose Tavus when a workflow needs a hosted conversational video presence. Text chat or a voice-only service may be sufficient when visual interaction adds no value.

Real-time video generation and the operated audiovisual pipeline require specialized models and serving capacity. A prompt or ordinary application script does not supply that infrastructure.

Suggested workflow

  1. Choose an appropriate stock or authorized face and define the conversation purpose. Keep the initial tool contract read-only and specify the expected result schema.
  2. Configure the correct environment and a short session limit before creating a conversation. Save the conversation ID and track the join and end states separately.
  3. Verify both sides of each tool call: the application handler and the response returned to the conversation. Add consequential actions only with the required authorization.

Acceptance checks before real work

  • Use the development MCP for configuration; the docs MCP is a separate search surface.
  • Check session termination and billing duration even when nobody joins.
  • A plausible spoken response must not substitute for a confirmed application result.

If the workflow fails

After a timeout, inspect the existing conversation before creating a replacement. End abandoned sessions and reconcile tool-call identifiers before retrying application writes.

Provider documentation

Explore services for adjacent needs

Continue building your toolkit