MCP client - inputSchema containing anyOf

[edeandrea]

I'm trying to build an MCP client to https://github.com/GongRzhe/Office-Word-MCP-Server. At startup, when ToolSpecificationHelper.toolSpecificationListFromMcpResponse runs, one of the entries in the array looks like the following.

You'll notice in the inputSchema.properties there are some properties that don't have a type key directly. Instead they have "anyOf": ["type":]...

Quarkus langchain4j seems to blow up with this, specifically in ToolSpecificationHelper.jsonNodeToJsonSchemaElement. It doesn't know what to do with anyOf.

{
  "name": "add_paragraph",
  "description": "\nAdd paragraph text to document\n\nParameters:\n- text: Paragraph text content\n- bold: Whether to bold\n- italic: Whether to italicize\n- underline: Whether to underline\n- font_size: Font size (points)\n- font_name: Font name\n- color: Text color (format: #FF0000)\n- alignment: Alignment (left, center, right, justify)\n",
  "inputSchema": {
    "properties": {
      "text": {
        "title": "Text",
        "type": "string"
      },
      "bold": {
        "default": false,
        "title": "Bold",
        "type": "boolean"
      },
      "italic": {
        "default": false,
        "title": "Italic",
        "type": "boolean"
      },
      "underline": {
        "default": false,
        "title": "Underline",
        "type": "boolean"
      },
      "font_size": {
        "anyOf": [
          {
            "type": "integer"
          },
          {
            "type": "null"
          }
        ],
        "default": null,
        "title": "Font Size"
      },
      "font_name": {
        "anyOf": [
          {
            "type": "string"
          },
          {
            "type": "null"
          }
        ],
        "default": null,
        "title": "Font Name"
      },
      "color": {
        "anyOf": [
          {
            "type": "string"
          },
          {
            "type": "null"
          }
        ],
        "default": null,
        "title": "Color"
      },
      "alignment": {
        "anyOf": [
          {
            "type": "string"
          },
          {
            "type": "null"
          }
        ],
        "default": null,
        "title": "Alignment"
      }
    },
    "required": [
      "text"
    ],
    "title": "add_paragraphArguments",
    "type": "object"
  }
}

[edeandrea]

I'm not sure if this is an issue here in Quarkus Langchain4j, or if its an issue with the MCP server, as I don't see the anyOf key anywhere when looking at the MCP spec for tools/list

[geoand]

cc @jmartisk @mkouba

[jmartisk]

What version are you using? I fixed that recently via langchain4j/langchain4j#2963

[jmartisk]

Yeah the MCP spec doesn't say anything about this, so I'm also a bit confused and I'm not sure if it's considered valid, but multiple MCP servers seem to be doing this, including @modelcontextprotocol/server-github which is kinda official

[edeandrea]

I'm using 0.26.2, but will probably switch to the 0.27 CR today ---------- Sent from a mobile device

…

On Thu, May 15, 2025, 12:52 AM Jan Martiska ***@***.***> wrote: *jmartisk* left a comment (quarkiverse/quarkus-langchain4j#1491) <#1491 (comment)> Yeah the MCP spec doesn't say anything about this, so I'm also a bit confused and I'm not sure if it's considered valid, but multiple MCP servers seem to be doing this, including @modelcontextprotocol/server-github which is kinda official — Reply to this email directly, view it on GitHub <#1491 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AACYXNY6SU4SGDNUZ3G7J6T26QMPBAVCNFSM6AAAAAB5D3SXKGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDQOBSGMYTAMRXHE> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[jmartisk]

0.27 CR2 is probably still too old, but I assume @geoand is going to release 1.0.0.CR1 which will contain langchain4j-bom 1.0.0, which contains langchain4j-mcp 1.0.0.Beta5, which fixes this (hopefully I didn't get too tangled in the versions)

[geoand]

That is correct. I'll do that tomorrow morning most likely

[edeandrea]

Then I will wait....