Skip to main content

MCP Clients

The Model Context Protocol (MCP) is an open standard for exposing tools to AI models. Many services ship an MCP server, and you can write your own. An AI MCP client connects MonoChat to one MCP server, so the tools that server offers can be given to your AI nodes and AI agents, next to the AI Tool functions you write in MonoChat.

You manage them in Settings › Automation › AI MCP Clients.

How it works​

  1. You add the MCP server's address and connect. MonoChat reads what the server offers: its Tools, Resources, Resource Templates and Prompts.
  2. When you save, MonoChat keeps that list as the client's contents.
  3. In a Create AI Chat Completion node or an AI Agent node you pick the tools you want that node to have, and you can rewrite each tool's description for the model.
  4. During the conversation the model decides when to call a tool; MonoChat calls the MCP server and returns the result to the model.

Only the tools you add to a node are available to it, so each node sees just what it needs.

Add an MCP client​

  1. Go to Settings › Automation › AI MCP Clients and click + (Create).

  2. Fill in the fields:

    FieldDescription
    NameA name for the server, shown next to each of its tools in the node drawers.
    DescriptionOptional notes.
    VersionThe client version MonoChat announces to the server when it connects, for example 1.0.0.
    EndpointThe base URL of the MCP server. MonoChat connects with the SSE transport at <Endpoint>/sse, so enter the address without /sse.
    Auth TypeNone, Basic (with Username and Password) or Bearer (with Token). The credentials are saved with the client.

  3. Click Connect. MonoChat connects to the server and shows what it found in tabs:

    • Server Info: the server's Name, Version, Protocol Version and Instructions.
    • Tools: each tool's name and description.
    • Resources, Resource Templates and Prompts, if the server offers any.

  4. If the list looks right, click Save. To change the address and try again, click the Disconnect button next to it.

tip

If Connect fails, check that the endpoint is reachable from the internet, that the server supports the SSE transport, and that the URL does not already end in /sse.

Update the tool list​

When the MCP server adds or changes tools, open the client with Update. MonoChat connects again and shows the current contents; click Update to save the new list.

Deleting a client also removes its contents, so nodes lose the tools that came from it.

Give tools to a node​

Create AI Chat Completion node​

  1. Open the node and scroll to AI MCP Client Content Tool Settings.
  2. Click Set AI MCP Client Content Tools. The AI MCP Client Content Tools drawer lists the tools already added to this node.
  3. Click + (Create), then choose a tool in AI MCP Client Content Tools. Tools are listed as <client name> - <tool name>.
  4. The Description is filled with the tool's own description. Edit it if you want to tell the model more precisely when to use the tool, for example "Use this to look up an order by its number". The description is required.
  5. Repeat for each tool, then save the node.

You can update the description of a tool or remove it from the node in the same drawer.

AI Agent node​

The AI Agent node has the same Set AI MCP Client Content Tools button in its Tool Settings section. The selected MCP tools are offered to the agent together with the node's AI scopes and AI tools, on every harness. See AI Agents.

info

An agent run checks that every MCP tool on the node still exists in your workspace. If you delete an MCP client that a node uses, runs of that node fail with "An AI scope or MCP tool was not found in this workspace." until you remove the tool from the node.

Permissions​

Access is controlled by the role permissions Read / Create / Update / Delete AI MCP clients in automation and Read AI MCP client contents in automation. See Roles.