-
-
Notifications
You must be signed in to change notification settings - Fork 605
Open
Description
When calling read_multiple_files through ChatGPT (Connectors, developer mode), the tool response surfaced to the model contains only metadata for each file.
Example (from ChatGPT tool inspector):
In contrast, calling the same tool on the same MCP server from another MCP client returns the metadata plus the full file contents, as expected per the spec:
Other tools like read_file do return full contents correctly in ChatGPT, so maybe ChatGPT expects the read_multiple_files response to follow a particular structure that differs from what DC currently emits, causing only the metadata portion to be surfaced?
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels