MCP server
@zentiq/mcp is a Model Context Protocol server that runs on your own machine and gives an AI assistant the Zentiq API as tools, with an API key you choose. Its tools are the hosted server's, made from the same operations as the reference by the same code: one per operation, named after its id (create_persona, list_portals). One more is only possible on your own machine: upload_document_file sends a file from your computer (PDF, Word, text, Markdown or CSV, up to 25 MB) to a knowledge pack.
For assistants that connect by signing in, with no key to copy, Zentiq also runs a hosted server: see Build with AI tools.
npx -y @zentiq/mcp@^1Coming to npm
@zentiq/mcp 1.0.0 is being released with the 2026-10-01 contract. The command above installs 1.0.0 or a later 1.x only, so it works as soon as 1.0.0 is out.Connect it#
Make a key for it in the console (More → API Keys), with only the scopes the assistant needs: the tools a key can't use are not offered. Then add the server to your assistant's MCP settings, as a command with the key in its environment:
{
"mcpServers": {
"zentiq": {
"command": "npx",
"args": ["-y", "@zentiq/mcp@^1"],
"env": { "ZENTIQ_API_KEY": "zq_live_…" }
}
}
}ZENTIQ_API_BASE points it at another address. It speaks MCP over standard input and output; its messages go to standard error.
What the tools may do#
- Every tool calls the API with your key: the key's scopes, the plan, the rate limits and the privacy rules all apply, as they do to your own code.
- A tool that deletes something, ends or starts a live conversation (starting one uses minutes), publishes, speaks into a live session, sends your data to another system, signs someone in, changes your people or candidates, or sends a file from your computer asks you first. Its first call changes nothing: it answers with what would change and a confirmation code. Only the same call again with that code does it, once, within five minutes, with exactly the same arguments.
- Text written by your website's visitors, customers or candidates, and your knowledge documents, comes back under
untrustedData, with a note that it is data, never instructions. - No tool makes or revokes API keys, records a hiring decision, or reviews a result: those stay with a person.
- A plan limit comes back as
upgrade_required, with the plan that includes it: the assistant tells you rather than trying again.