Tool Input Data Types

[zthafir]

I’m working on a case where I want to use an MCP server (specifically a custom MCP server). This server contains a group of tools, and each tool may accept different input arguments with different data types.

The issue I’m facing is that whenever I try to use a tool, Flowise always converts the input arguments into strings, which then causes errors.

For example, when using the Firecrawl MCP server and selecting the FIRECRAWL_CRAWL tool , I get the following error:

[{"type":"text","text":"Error: Request failed with status code 400. 
Error: Bad Request - [
  {"code":"invalid_type","expected":"number","received":"string","path":["maxDepth"],"message":"Expected number, received string"},
  {"code":"invalid_type","expected":"number","received":"string","path":["limit"],"message":"Expected number, received string"},
  {"code":"invalid_type","expected":"boolean","received":"string","path":["ignoreQueryParameters"],"message":"Expected boolean, received string"}
]"}]

Here are the screenshots of the Flow setup that produces this issue:

So my question is:
Is there a way to prevent Flowise from converting all arguments to strings, or to handle this type casting properly so the tools receive the correct types (number, boolean, etc.)?

[Ganesh-403]

@zthafir
This looks like a type conversion issue in the current MCP integration rather than something specific to the Firecrawl MCP server.

The error:

Expected number, received string
Expected boolean, received string

indicates that the MCP server is correctly validating its input schema, but Flowise is sending values like:

{
  "maxDepth": "3",
  "limit": "10",
  "ignoreQueryParameters": "true"
}

instead of:

{
  "maxDepth": 3,
  "limit": 10,
  "ignoreQueryParameters": true
}

Since the Firecrawl MCP server uses strict schema validation, it rejects those requests with a 400 Bad Request. :contentReference[oaicite:0]{index=0}

From what I can tell, I couldn't find any setting in Flowise's Custom MCP node that lets you specify or override the runtime data types for tool arguments. The documentation explains how to configure the MCP server and select available actions, but it doesn't mention automatic type coercion for tool inputs. :contentReference[oaicite:1]{index=1}

There have also been related reports where MCP tools failed because Flowise passed values in an unexpected type (for example, sending an object as a string), which suggests this area has had interoperability issues with strongly typed MCP servers. :contentReference[oaicite:2]{index=2}

Things you can try

1. Let the LLM populate the tool arguments

   • If you're manually entering tool input values, Flowise may treat them as strings.
   • When possible, allow the agent to generate the arguments based on the MCP tool schema.

2. If you control the MCP server, add type coercion

   • For example, convert "3" → 3 and "true" → true before validation.
   • This makes the server more tolerant of clients that serialize everything as strings.

3. Test with another MCP client

   • If the same tool works correctly in clients such as Claude Desktop or Cursor using the same server, but fails only in Flowise, that would strongly indicate the issue is in Flowise's MCP client rather than the MCP server itself.

At this point, I suspect this is a limitation or bug in Flowise's handling of MCP tool arguments rather than a configuration issue.

A couple of questions that would help confirm that:

• Which Flowise version are you using?
• Are these values being entered manually in a Tool node, or is an Agent generating the tool call?
• Does the same MCP tool work correctly from another MCP client (e.g. Claude Desktop, Cursor, or the MCP Inspector)?

If it reproduces across different MCP servers that expect typed parameters, I'd recommend opening a GitHub Issue, as Flowise should ideally preserve the JSON schema types (number, boolean, object, etc.) when invoking MCP tools.

If this solves your problem, feel free to mark it as the accepted answer so others can find it easily.