issue: Reference MCP Server (Sequential Thinking) not compatible with mcpo

[enving]

Check Existing Issues

• I have searched the existing issues and discussions.
• I am using the latest version of mcpo.

mcpo Version

v0.0.12

Open WebUI Version (if applicable)

v0.6.5

Operating System

Ubuntu 24.04

Browser (if applicable)

Chrome

Confirmation

• I have read and followed all instructions in README.md.
• I am using the latest version of both MCPO and Open WebUI.
• I have included the browser console logs.
• I have included the Docker container logs.
• I have listed steps to reproduce the bug in detail.

Expected Behavior

The integration of the Sequential Thinking MCP server with OpenWebUI should work correctly. When a user makes a request that requires sequential thinking, OpenWebUI should be able to call the /sequentialthinking/sequentialthinking endpoint with properly typed JSON parameters, and the tool should perform a step-by-step analysis.

Actual Behavior

The integration of the Sequential Thinking MCP server with OpenWebUI should work correctly. When a user makes a request that requires sequential thinking, OpenWebUI should be able to call the http://mcpo:8000/sequentialthinking endpoint with properly typed JSON parameters, and the tool should perform a step-by-step analysis.

We get a error in the container logs: ""Input should be a valid integer, unable to parse string as an integer"

The problem appears to be a data type incompatibility issue between OpenWebUI and the Sequential Thinking server. OpenWebUI sends parameters as strings (also for null values and booleans), while the API expects concrete types (integer, boolean, null).

Steps to Reproduce

1. Create a Docker Compose setup with OpenWebUI and MCPO Server

2. Configure Sequential Thinking in the config.json:

   "sequentialthinking": {
   "command": "npx",
   "args": [
   "-y",
   "@modelcontextprotocol/server-sequential-thinking"
   ]
   }

3. Configure in Openwebui:

   • URL: http://mcpo:8000/sequentialthinking
   • Auth: Bearer Token with API Key

4. Activate native function calling to see the status in the chat fenster or leave it off and see in container logs

5. In a chat, ask a question that requires sequential thinking, e.g., "Analyze step by step the business model of 1komma5.com"

6. The MCPO logs show a 422 error

The manual curl command works fine:

curl -X POST "http://localhost:8000/sequentialthinking/sequentialthinking"
-H "Content-Type: application/json"
-H "Authorization: Bearer ..."
-d '{
"thought": "I am analyzing the business model of 1komma5.com",
"nextThoughtNeeded": true,
"thoughtNumber": 1,
"totalThoughts": 5
}'

But the request generated by OpenWebUI fails.

Logs & Screenshots

Faulty request from OpenWebUI to the Sequential Thinking server:

Error Message in MCPO container logs:

docker logs mcpo

{
"error": "HTTP error 422: {"detail":[{"type":"int_parsing","loc":["body","revisesThought"],"msg":"Input should be a valid integer, unable to parse string as an integer","input":"null"},{"type":"int_parsing","loc":["body","branchFromThought"],"msg":"Input should be a valid integer, unable to parse string as an integer","input":"null"}]}"
}

Error message in UI:

{
"thought": "Let's break down the business model of 1komma5.com into its key components and analyze each step in detail.",
"next_thought_needed": "True",
"thought_number": "1",
"total_thoughts": "5",
"is_revision": "False",
"revises_thought": "null",
"branch_from_thought": "null",
"branch_id": "null",
"needs_more_thoughts": "False"
}
{
"error": "HTTP error 422: {"detail":[{"type":"missing","loc":["body","nextThoughtNeeded"],"msg":"Field required","input":{"thought":"Let's break down the business model of 1komma5.com into its key components and analyze each step in detail."}},{"type":"missing","loc":["body","thoughtNumber"],"msg":"Field required","input":{"thought":"Let's break down the business model of 1komma5.com into its key components and analyze each step in detail."}},{"type":"missing","loc":["body","totalThoughts"],"msg":"Field required","input":{"thought":"Let's break down the business model of 1komma5.com into its key components and analyze each step in detail."}}]}"
}

Additional Information

No response

[EntropyYue]

This is due to the model's output tool call not using the correct format; all numbers and booleans should not be strings. Using a model with better tool call capabilities can solve this issue, for example, Qwen3-14B

[iansmathew]

@EntropyYue I am running Qwen3-30B-Thinking and it calls the sequential thinking MCP correctly.

However, It only performs the tool call once and presents an output. The next thought only occurs after the user inputs something else like "continue".

The expected behavior would be for the model to make multiple calls to the SequentialThinking MCP and present a final output after all thoughts are completed.

[iansmathew]

Nevermind, it seems to be invoking the tool currently when the function calling is set to Native.

[EntropyYue]

This seems to be due to the activation parameters of this model being too small; Try its instruct version, and I tested Qwen3-8B(non-reasoning mode), it works