Unable to integrate Mongodb mcp with google adk

[Ateeb626]

Describe the bug
When using MCPToolset with MongoDB MCP in an LlmAgent, some function parameters have enum values that are integers (e.g., [256, 512, 1024, ...]) while the schema specifies type: string. This causes google.genai.errors.ClientError: 400 INVALID_ARGUMENT when the agent is loaded or tools are called, with messages like:

Invalid value at 'tools[0].function_declarations[13].parameters.properties[2].value.items.any_of[1].properties[0].value.properties[7].value.properties[0].value.enum[0]' (TYPE_STRING), 256

To Reproduce
Steps to reproduce the behavior:

1. Install the latest google-adk and 'MongoDB mcp server'
2. Start the MongoDB MCP server before running the agent:

npx -y mongodb-mcp-server \
  --connectionString "mongodb://localhost:27018?directConnection=true" \
  --transport http \
  --httpHost=0.0.0.0 \
  --httpPort=8080

3. Create an agent using MongoDB MCP:

from google.adk.agents import LlmAgent
from google.adk.tools.mcp_tool import MCPToolset, StreamableHTTPConnectionParams

mcp_server_url = "http://127.0.0.1:8080/mcp"

root_agent = LlmAgent(
    model="gemini-2.5-flash",
    name="mongo_mcp_agent",
    instruction="Help with MongoDB data access",
    tools=[MCPToolset(connection_params=StreamableHTTPConnectionParams(url=mcp_server_url))]
)

4. Run adk web . from the agent folder.
5. Observe errors related to TYPE_STRING and enum integer values.

Expected behavior
Integer enum values returned by MongoDB MCP should either:

• Be automatically cast to strings when the schema type is string, or
• The schema itself should use type: integer, so ADK can validate properly.

Desktop

• OS: macOS 13.5
• Python version: 3.13
• ADK version: [latest]

Model Information:

• Using LiteLLM: No
• Model: gemini-2.5-flash

Additional context
This breaks integration of MongoDB MCP with LlmAgent in ADK. Guidance on proper schema handling or automatic enum type normalization is needed.

[surajksharma07]

@Ateeb626 The issue is MongoDB MCP's schema has type: "string" with integer enum values [256, 512...], which Gemini rejects.
As a workaround, create custom Python tool functions for MongoDB operations instead of using MCPToolset - this gives you control to convert params to proper types.
We're looking more into this to auto-normalize enum values in MCP schemas.
Thanks!

[Ateeb626]

Is this not an issue in how Gemini interprets the enum of the JSON Schema. JSON Schema supports numeric types in an enum, and even a mix of types which is the response that the MongoDB MCP returns.

[surajksharma07]

You're right @Ateeb626 - JSON Schema spec does support mixed types in enums,so MongoDB MCP isn't technically wrong. The issue is Gemini API has stricter validation than the JSON Schema spec - it requires enum values to match the declared type exactly.
ADK needs to bridge this gap by normalizing MCP schemas to fit Gemini's constraints.
We're tracking this as an ADK enhancement to auto-convert enum values to match the declared type during schema translation.
For now, the custom Python tool workaround is the fastest path forward.
Thanks!

[surajksharma07]

@DeanChensj Please have a look into it.