Agents limited to answering questions can’t take real-world action, leaving teams stuck with manual steps, wasted time, and disconnected workflows.
Easily connect to MCP (Model Context Protocol), a universal adaptor that connects to thousands of tools and databases. With this, you can perform 1000s of context-aware custom actions—book calls, check orders, update records, or trigger workflows — securely and seamlessly, while staying grounded in your knowledge.
Execute tasks confidently while staying grounded in accurate knowledge
Maintain data security and business accuracy with fast, frictionless action
Connect to 1000s of systems with no limits through seamless integrations
Simply copy-paste the URL of any remote MCP server, and watch your agent transform from an information provider into a problem solver. No complex integrations. No months of development. Just real actions, real results, in minutes.
"We love that we can trigger actions directly from our Agents—and trust the outcome!"
Learn more about Custom Actions in our docs.
An MCP (Model Context Protocol) server is a standardized bridge that connects AI agents to external tools and data sources. Instead of building custom integrations for each AI platform, MCP provides a universal protocol that works across different AI applications, including CustomGPT.ai.
The key benefit is modularity and reusability – one MCP server can provide tools for database queries, API interactions, file access, or specialized computations, and any MCP-compatible AI can use it. When connected, the AI sees these tools, understands how to use them, and can call them as needed during conversations. This keeps the AI focused on reasoning while delegating specific actions to purpose-built tools, making it easy to extend AI capabilities without modifying the AI application itself.
A knowledge‑base query retrieves or synthesizes information that already exists inside CustomGPT.ai.
A Custom Action is user‑defined code that executes outside CustomGPT.ai – creating, updating, or retrieving live data in your own systems. In short, KB = read‑only insight; Action = read‑write execution.
Most platforms (ChatGPT, Claude, Zapier Actions, N8N, LangChain Tools) let you connect MCP server to their agent and use it within your organizations. CustomGPT.ai Custom Actions go one step further: you can package the agent itself – complete with its actions – and let your customers or website visitors use it. For example, embed an agent on your site so prospects can both chat and book a call with you in a single conversation, all powered by your MCP server.
Navigate to Agent › Actions. The page lists existing actions and an Add Action button for new ones.
At launch we support:
Yes. After registering an action you can chat with the agent in the app using test prompts before exposing it to end‑users.
Initially, one agent → one MCP server. Multi‑server support is on our roadmap.
The agent selects the appropriate action, but it pauses and asks the user for approval (summary of intent + inputs). You can optionally disable this confirmation for a fully autonomous flow.
All data that passes through CustomGPT.ai remains covered by our SOC 2 Type II controls and GDPR tooling. Anything processed only on your MCP server is your responsibility to secure and audit.
If the agent detects a failed or suspicious response it will gracefully decline: “Sorry, I can’t complete that action right now.” We also enforce reasonable timeouts and retry caps.
Yes. Each call is written to the Action Logs tab with timestamp, user, parameters, response, and status.
No. Each action call counts as two standard queries – one to decide and one to process the result.
