Fix Gemini schema errors with LangChain.js + MCP

Drop-in replacement that unblocks MCP tool schemas in Gemini

This library provides a drop-in replacement for ChatGoogle from @langchain/google that fixes Gemini function-calling schema errors when using LangChain.js with MCP servers. It automatically transforms MCP tool schemas with unsupported constructs into Gemini-compatible function-calling schemas at tool binding time.

The schema error usually appears as either a Gemini API request error:

RequestError: Invalid JSON payload received.
Unknown name "exclusiveMaximum" ...
Unknown name "exclusiveMinimum" ...
Copy

or a LangChain-side schema validation error before the request is sent:

InvalidInputError: Gemini does not support union types in function schemas.
Use a single type instead.
Copy

This commonly appears when tools from MultiServerMCPClient include schemas that are valid JSON Schema but outside Gemini's supported OpenAPI-like subset.

How to Use This Library

Replace:

import { ChatGoogle } from "@langchain/google/node";

const model = new ChatGoogle({ model: "gemini-2.5-flash" });
Copy

with:

import { ChatGoogleEx } from "@h1deya/langchain-google-ex";

const model = new ChatGoogleEx({ model: "gemini-2.5-flash" });
Copy

That's it. Keep passing the original MCP tools to LangChain:

const mcpTools = await client.getTools();
const agent = createAgent({ model, tools: mcpTools });
Copy

ChatGoogleEx transforms the tool schemas inside bindTools(), preserving the original LangChain tool objects and their execution behavior.

When using a Google AI Studio / Gemini Developer API key, pass it explicitly:

const model = new ChatGoogleEx({
  model: "gemini-2.5-flash",
  apiKey: process.env.GOOGLE_API_KEY,
});
Copy

This avoids accidentally falling back to Vertex AI / Google Cloud authentication when the environment is not configured the way @langchain/google expects.

Prerequisites

• Node.js 20+
• Google API key configured for @langchain/google
• @langchain/google, langchain, and @langchain/mcp-adapters
• MCP servers you want to use

Tested with @langchain/google@0.2.3, langchain@1.5.10, and @langchain/mcp-adapters@1.1.4.

Installation

npm i @h1deya/langchain-google-ex @langchain/google langchain @langchain/mcp-adapters
Copy

Complete Usage Example

import "dotenv/config";
import { ChatGoogleEx } from "@h1deya/langchain-google-ex";
import { createAgent } from "langchain";
import { MultiServerMCPClient } from "@langchain/mcp-adapters";

const client = new MultiServerMCPClient({
  throwOnLoadError: true,
  useStandardContentBlocks: true,
  mcpServers: {
    fetch: {
      transport: "stdio",
      command: "uvx",
      args: ["--with", "mcp<2", "mcp-server-fetch==2025.4.7"],
    },
  },
});

try {
  const mcpTools = await client.getTools();
  const model = new ChatGoogleEx({
    model: "gemini-2.5-flash",
    apiKey: process.env.GOOGLE_API_KEY,
  });
  const agent = createAgent({ model, tools: mcpTools });

  const result = await agent.invoke({
    messages: [
      {
        role: "user",
        content: "Fetch the raw HTML content from bbc.com and tell me the title",
      },
    ],
  });

  console.log(result.messages.at(-1)?.content);
} finally {
  await client.close();
}
Copy

Why This Happens

Gemini's function-calling schema format accepts only a subset of JSON Schema. Some MCP servers publish schemas containing fields such as exclusiveMinimum, exclusiveMaximum, propertyNames, additionalProperties, or complex union shapes. Those schemas can be rejected before any tool call runs.

@langchain/google already performs some schema normalization, but current versions still pass through schema keywords that Gemini rejects in MCP tool definitions. ChatGoogleEx adds a compatibility layer for those cases.

MCP servers that have shown this kind of issue include:

• airtable-mcp-server
• mcp-server-fetch==2025.4.7
• GitHub Copilot MCP server
• @notionhq/notion-mcp-server

In local integration tests, simple schemas such as a weather MCP server worked with both ChatGoogle and ChatGoogleEx. More complex schemas from Fetch, Airtable, and GitHub failed with ChatGoogle and succeeded with ChatGoogleEx.

Debugging: Verbose Logging

Set this environment variable to see schema transformations:

LANGCHAIN_GOOGLE_EX_VERBOSE=true
Copy

Example output:

Transforming 3 MCP tool(s) for Gemini compatibility...
fetch: 2 exclusive bound(s) converted, 1 unsupported format(s) removed
Summary: 1/3 tool(s) required schema transformation
Copy

Features

• Drop-in replacement for ChatGoogle
• Automatic MCP tool schema transformation at bindTools() time
• Preserves original LangChain tool objects
• Converts type arrays such as ["string", "null"] to nullable schemas
• Filters invalid required fields
• Removes or converts unsupported JSON Schema keywords
• Resolves local $ref, $defs, and definitions where possible

Known Limitations

• Unresolved references are simplified to generic object schemas.
• Tuple-style arrays keep only the first item schema.
• Non-string enum values are dropped.
• Complex oneOf / allOf schemas may be simplified, which can loosen validation.

These adjustments keep most MCP tools working, but rare edge cases could behave differently from the original schema. Please report issues at GitHub Issues.

See DESIGN_DECISIONS.md for implementation details.

API Reference

API Reference

Change Log

CHANGELOG.md

License

MIT