A growing share of Quo integrations are written by agents or run as agents. The docs, the spec, and the error messages are all written with that reader in mind.
An AI shows up here in two ways, and they need different things.
An agent that uses Quo
You don’t need this API to put an AI to work on your phone system. Quo ships an official MCP (Model Context Protocol) connector, available on Starter, Business, and Scale plans, that gives Claude and ChatGPT direct access to your workspace.
Once connected, you can work in plain language across your real communication history:
- “What are the top reasons prospects say no after a sales call?”
- “Find customers with appointments this week and send each a reminder with the correct details.”
- “Update the contacts mentioned in today’s call notes with their new titles and numbers.”
To connect Claude: Settings → Connectors → Browse connectors → Quo → Connect. Details and the ChatGPT setup live in the Quo Resource Center, or find Quo in the Claude connector directory.
The connector authenticates through your Quo account and acts with your access. The example prompts above are real capabilities, including sending messages, so extend it the trust you’d extend any teammate who can text your customers.
An agent that builds on Quo
If an LLM is writing your integration, whether in a code editor or an autonomous loop, feed it the same sources of truth we maintain for humans:
| Resource | What it’s for |
|---|
llms.txt | A structured index of these docs. The right starting point for any agent. |
llms-full.txt | The entire documentation set as one plain-text file, for large context windows. |
| OpenAPI specification | The exact contract for version 2026-03-30: every endpoint, parameter, and schema. The ground truth for codegen and validation. |
| Docs bundle (zip) | The documentation packaged for offline or air-gapped agent setups. |
llms.txt and the spec can answer almost any “how do I…?” about this API without guessing. When its output disagrees with the spec, the spec is right.
Ground rules
These are the practices that separate agent-built integrations that run quietly for years from the ones that break in production.
- The key never enters the conversation. Your API key belongs in an environment variable or the connector’s configuration. Never put it in a prompt or in generated source code. Anything an LLM reads, assume it may repeat.
- Give the agent its own key. Name it (
claude-agent, gpt-ops) so you can see its traffic, and revoke it without touching anything else.
- Tell it the budget. State the 10 requests per second limit up front. Left to its own devices, an agent will poll.
- Check generated code against the spec. Code that compiles can still call endpoints that don’t exist or send parameters the API ignores. The OpenAPI spec is machine-checkable; use it in review, or have the agent use it itself.
- Let the errors do their job. The API returns structured, field-level errors precisely so an agent can read
errors[].path, fix the request, and continue. Surface them to the agent raw instead of swallowing them.
Choosing your interface
There are two ways to put an AI in front of Quo today, and they solve different problems.
| You want to… | Use | In a sentence |
|---|
| Ask questions and take actions in plain language | MCP connector | An agent works your workspace conversationally, with your permissions. |
| Read and write on your own schedule | REST API | Your code (or your agent’s code) pulls exactly what it needs, versioned and paginated. |
