MCP connector availability

What it is

MCP server is the sixth Call Action type. MCP — the Model Context Protocol — is an open standard for exposing tools to an AI model: you run an MCP server, it advertises the tools it offers, and a client discovers and calls them.

In Agentency the type reads "Connect a remote MCP server's tools." It is availability-dependent and switched off by default. Unless whoever runs your Agentency deployment has turned it on, you should treat MCP as unavailable and build with the other five types.

This page exists so you can tell, in under a minute, whether MCP is on for you — and get on with a working alternative if it is not.

When you would use it

Read this if you saw MCP in the type picker and want to know whether to plan around it, or if you already run an MCP server and are trying to work out why Agentency will not list its tools.

Where to find it

Open Dashboard → Chatbots → your chatbot → Actions → New action. If MCP is available, MCP server appears as one of the tiles under Start from scratch.

How to tell whether it is on

1. Open the New action chooser.
2. Look for an MCP server tile under Start from scratch. If the tile is not there, MCP is off for your deployment. Stop here.
3. If the tile is there, select it and enter your MCP server URL.
4. Select Discover tools. This is the real test.
5. A working setup returns a list — "3 tools found" — and you pick the tool to expose.
6. A message saying MCP is not enabled means the feature is off however correct your server is.
7. Do not promise MCP to a customer until you have seen a successful discovery on your own workspace.

The type picker is populated from what your deployment actually supports, so the absence of the tile is a reliable answer. Discovery is the definitive one.

What it would look like when it is on

For completeness, the configuration is:

• MCP server URL — the address of your server, reachable over public HTTPS.
• Auth token — an optional bearer token, stored write-only like every other secret.
• Discover tools — queries the server and lists what it exposes.
• Tool name — the one tool this action exposes. One action, one tool; a server with eight tools you want all of is eight actions.
• Read-only tool (skip confirmation) — MCP tools require the visitor to confirm before running by default. Turn this on only for a tool that reads data and never changes anything.

Parameters are taken from the tool's own schema at discovery time, so you do not hand-write them.

The same protections that apply to a Custom API action apply here: private and internal addresses are refused, calls are time-boxed, responses are size-capped, repeated failures pause the endpoint briefly, and the server's response is treated as untrusted text rather than as instructions.

What to use instead

Almost everything people want MCP for is already reachable through Custom API, which is on by default:

You want	Use
Call one endpoint on your own API	HTTP connectors
A named service — Zendesk, Shopify, Slack, Notion…	Use a preset
Something with no template and no simple API	The Zapier or Make bridge in the preset reference
Many endpoints at once	Import an OpenAPI document — see Create a Call Action
Escalate to a person	Hand off to a human

The OpenAPI import is the closest thing to MCP's appeal: paste your API document as JSON and Agentency seeds an action per operation for you to review and save. It gets you the "expose my whole toolset" outcome without the protocol.

What you will see

If MCP is off, the type simply is not offered, and an attempt to discover tools reports that MCP is not enabled rather than pretending the server was unreachable. That distinction matters when you are debugging: not enabled is about Agentency, unreachable is about your server.

Any MCP actions that already exist on a chatbot stay in the library and are visible, but are not offered to the model while the type is unavailable.

Limits and plan notes

• Call actions as a whole are available by default; MCP specifically defaults off. They are two different switches.
• MCP is not a plan feature you can buy. It is a deployment-level capability, so no upgrade turns it on. See Plans and what you get.
• If it is enabled, an MCP action counts as one of your enabled Call Actions like any other type: Free 0, Starter 3, Standard 8, Pro 12, Agency 20 per chatbot.
• MCP tools require confirmation by default. Only mark a tool read-only when you are certain it changes nothing.
• Enabling MCP later does not disturb your existing actions. The types sit side by side.
• MCP is unrelated to messaging channels that are also marked coming soon. See Coming soon channels.

Common problems

There is no MCP tile in the chooser.

MCP is off for your deployment. This is not a dashboard toggle and not something an upgrade changes. Use Custom API instead.

Discovery says MCP is not enabled.

Same answer, arriving from a different direction. Nothing about your server is at fault.

Discovery returns no tools, but the server is reachable.

MCP is on and Agentency reached your server, but it advertised nothing. Check the URL points at the MCP endpoint rather than a landing page, and that the token has permission to list tools.

Discovery fails to reach my server.

The address must be reachable over the public internet. A server on a private network, a loopback address, or an internal hostname is refused for the same reason Custom API actions are.

I built an MCP server and Agentency will not use it.

If the type is unavailable, that is the whole answer, and there is nothing to configure your way around. Expose the same capability as a small HTTPS API and use a Custom API action — the visitor-facing result is identical.

Do I need MCP to launch?

No. A live chatbot needs knowledge, a welcome message, and an active hosted link or embedded widget. See Go-live checklist.