`set_function_prototype` always fails, loses prototype parameter

[laanwj]

set_function_prototype MCP tool does not work. The Binary Ninja plugin server receives HTTP requests without the prototype query parameter, even though the MCP call provides it.

MCP tool invocation (what we send):

{
  "name_or_address": "TIMERA_BaseInit_packed",
  "prototype": "int32_t TIMERA_BaseInit_packed(void* timer, void const* cfg)"
}

Expected HTTP request:

GET /setFunctionPrototype?prototype=int32_t%20TIMERA_BaseInit_packed(void*%20timer%2C%20void%20const*%20cfg)&name=TIMERA_BaseInit_packed

Actual HTTP request:

"GET /setFunctionPrototype?name=TIMERA_BaseInit_packed HTTP/1.1" 400 -

It misses the prototype parameter, resulting in t always returns 400. This may be an escaping issue at some level of the stack causing the parameter to be ignored, i'm not sure. It does happen even with simple prototypes like void x(void). The other MCP tools provided by binary ninja seem to work fine.

This is with opencode and Deepseek V4 Pro (though i'm not sure that matters).

Edit: i've narrowed it down to a problem in the Axios dependency. Axios 1.13.2 includes prototype in the URL, but Axios 1.16.0 drops it. So it appears there's been a regression there.

[laanwj]

The breaking change is in Axios v1.13.5 (released Feb 8, 2026), PR #7369:

Prototype Pollution (DoS): Hardened mergeConfig to ignore __proto__, constructor, and prototype keys, preventing denial-of-service via prototype pollution when merging user-supplied config.

Axios now strips any key named prototype from objects that pass through its config merge path, including the params object. That's why:

• Axios 1.13.2 (local): GET /setFunctionPrototype?prototype=int32_t+main(void)&name=main: ok
• Axios 1.16.0 (npm): GET /setFunctionPrototype?name=main: prototype silently dropped

A potential fix would be to rename the HTTP query parameter key. The server already accepts fallbacks at http_server.py:1313:

proto = params.get("prototype") or params.get("signature") or params.get("type")