Skip to main content

​
Filtering Irrelevant Fields with JMESPath

When an MCP action runs successfully, external APIs invariably return metadata exclusively intended for graphical user interfaces (GUIs), such as pagination coordinates (next_page_cursor), styling data (font_color), or HTML attributes. You can strip this structural noise instantly using JMESPath projection.

​
Example: Stripping Root Metadata

The Raw Upstream API Response: 
{
 "_metadata": {
 "latency_ms": 142,
 "has_more": false,
 "theme": "dark"
 },
 "customers": [
 { "id": 1, "name": "Google", "status": "active" },
 { "id": 2, "name": "Apple", "status": "pending" }
 ]
}
Instead of sending those useless _metadata tokens to Claude or OpenAI through HasMCP, simply provide the following JMESPath execution string in the Provider Tool settings: The JMESPath Input: customers The Final Pruned Output Delivered to the LLM: 
[
 { "id": 1, "name": "Google", "status": "active" },
 { "id": 2, "name": "Apple", "status": "pending" }
]
By simply addressing the root object customers, the engine immediately discards everything surrounding it automatically.