1. Start with read-only tools
Do not let early tests create, delete, send, purchase or modify real data. Use deterministic read-only tools first.
MCP and tool calling
Test agent tool schemas through an OpenAI-compatible route
MCP servers and OpenAI-compatible tool calls both depend on clear schemas. Use TKEN for gateway-level smoke tests when your model route supports the tool-calling behavior your agent requires.
Tool schema smoke test
chat completions
{
"model": "MiniMax-M2.7",
"messages": [
{"role": "user", "content": "Check order status for order_123."}
],
"tools": [
{
"type": "function",
"function": {
"name": "lookup_order",
"parameters": {
"type": "object",
"properties": {
"order_id": {"type": "string"}
},
"required": ["order_id"]
}
}
}
]
}Validate JSON schemas before sending traffic
Test model-specific tool behavior
Redact logs before debugging or sharing
Compatibility test plan
Tool calling is not only an endpoint question. You need to verify model support, schema adherence, retry behavior, timeout handling and how your app executes or rejects a proposed tool call.
Gateway
https://www.tken.shop/v1
Schema type
JSON Schema
Test first
One harmless read-only tool
Risk control
Backend executes tools, not the model
Agent safety checklist
2. Validate every tool argument
Validate generated arguments against the expected schema before your backend executes a tool call.
3. Add fallback behavior
Decide whether to retry, ask the user, switch model routes or stop when a tool call is missing, malformed or unsafe.
TKEN is an independent third-party API gateway. It is not an official
MCP, OpenAI, Anthropic, Google, DeepSeek, Qwen, Kimi, MiniMax or GLM
endpoint.
Run one safe tool-calling smoke test
Use a harmless schema, confirm the returned tool call shape and only then test more complex agent workflows.