Skip to main content

Documentation Index

Fetch the complete documentation index at: https://agnt.mintlify.app/llms.txt

Use this file to discover all available pages before exploring further.

agntdata Agents lets you ship a working Claude agent in minutes — no infrastructure, no glue code. Every agent runs on Anthropic’s managed runtime and pulls its tools from the agntdata MCP gateway, which exposes:
  • Data APIs — every API in your account (LinkedIn, X, YouTube, etc.) is callable as a typed tool.
  • Integrations — anything you’ve authenticated with an API key (Gmail, HubSpot, …) becomes an authenticated tool the agent can invoke on your behalf.
  • Connectors — first-party OAuth installs you complete in the dashboard (Slack today; Gmail/Calendar/Notion next). Tools like slack_post_message show up here.
  • Utilities — built-in tools like credits_balance, webhook_send, and escalate_to_human.

​
Two surfaces

The Builder

A meta-agent that interviews you in natural language and configures the deployed agent for you. It maintains long-term memory of your design decisions across sessions.

Deployed Agents

Production-ready Claude agents you can chat with from the dashboard, embed via API, or wire into Slack DMs and channels.

​
How it works

  1. Open Agents → Build with AI and describe what you want.
  2. The Builder proposes a configuration (system prompt + tool list). You approve it.
  3. agntdata mints an agent-scoped API key, registers the agent with Anthropic, and exposes an MCP gateway URL the runtime uses to fetch tools.
  4. Your agent is live. Talk to it from the dashboard, the API, or Slack.

​
Concepts

  • System prompt — the agent’s persistent role and rules.
  • Tools — discrete capabilities the agent can call. Each invocation is logged to usage_logs and (for billable APIs) deducted from your workspace credits.
  • Sessions — every conversation gets an agent_session row with the full transcript so you can audit what the agent did.
  • Permissions — mark sensitive tools as always_ask to require human approval before the agent uses them.
  • Escalations — agents can ask a human via escalate_to_human. The question lands in your Inbox (and a Slack channel, if connected).
See Slack connector for routing inbound messages to your agents and Permissions for human-in-the-loop control.