For reference see: https://discord.com/channels/1211482211119796234/1379841273002725498 @Shubham Bajaj Thanks, adding the MCP in the global tools now seems to work. However 1. The documentation is still incorrect and not useful at all: https://docs.vapi.ai/tools/mcp 2. The method shown in the mcp docs, by adding the tool to the

tools

object in the

model

config does not work. Just to be clear this example is broken.

ts
{
  "model": {
    "provider": "openai",
    "model": "gpt-4o",
    "messages": [
      {
        "role": "system",
        "content": "... removed for brievety"
      }
    ],
    "tools": [
      {
        "type": "mcp",
        "name": "mcpTools", // this field does not exist in the api, however a name is required
        "server": {
            "url": "https://actions.zapier.com/mcp/actions/"
        }
      }
    ]
  }
}

It will result in this error:

ts
Body: {
  "message": [
    "model.each value in tools.property name should not exist"
  ],
  "error": "Bad Request",
  "statusCode": 400
}

Adding the name in the

function.name

parameter allow us to save the configuration, however then the assistant fails to add and execute the mcp tools, it results in the same issue i described before, where it will literally try to call that function name on the mcp server. So it seems like the MCP tool in the

model.tools

config are interpreted as regular tools, instead of being used to fetch the mcp tools and dynamically add them to the assistants tools.

You're correct—the current documentation example for adding MCP tools to

model.tools

is broken and will result in a validation error. The only reliable workaround at this time is to add the MCP tool in the global tools section of your assistant configuration, not within

model.tools

. This ensures dynamic tool injection works as expected. The documentation does not yet reflect this, so for now, avoid placing MCP tools in

model.tools

and use the global tools configuration instead. If you need further help, consider reaching out on the [Vapi Discord support channel](https://discord.gg/pUFNcf2WmH). Source: - [Vapi MCP Tool Documentation](https://docs.vapi.ai/tools/mcp)

Hey Elco! To help track down this issue, could you share: - The call ID This would really help us figure out what went wrong!

it's the same error as the call id's i already shared in the other closed issue. Without name in the tool: 3e8fb647-58d5-4b48-80eb-f09a343f5b85 With name in the tool: 5767c2c1-5abf-48ea-890f-1920e2c0ff62 However we'll just resort to using global tools for mcp like vapi support bot suggests. Please get it corrected in the docs.

Hey there, thanks for bringing this issue to my attention. I've verified the situation. It seems that when you use the [assistant.model.tools](http://assistant.model.tools), the MCP child tools that are fetched and populated aren't being correctly added as tools to the required LLM model, which is causing the broken experience for you. In contrast, when you add tools at the global level, they are being correctly populated into the required LLM model or style. I've reported this issue to the team, and they're working on it. You can track the progress of this issue from here once it moves out of the backlog and into the start stage. vapi.steelsync.io/public-support-board