Repository navigation
Reconsider ambiguous implementation note regarding missing inputSchema in MCP headers #2974
Description
What's broken?
The spec is ambiguous or self-contradictory
Where in the spec or docs?
What should happen?
We should evaluate removing the current implementation note regarding the handling of mcp-param headers when the inputSchema is not available. (the one edited in #2972)
The current language suggests a scenario where a client might make a tool call without having retrieved the inputSchema, which is unclear and likely represents an invalid state that shouldn't be explicitly handled in this manner. It is generally expected that clients will have fetched the tools/list information prior to executing calls.
Proposed Action:
Delete the implementation note regarding the inputSchema condition entirely.
Although we need to make sure we do explain that if you get rejected, you should tools/list to get the schema for custom header info, and then retry.
What actually happens?
...
Anything else?
No response
Activity
piyushbag commentedon Jun 28, 2026
Planning a small docs follow-up for this issue: remove the
inputSchemaavailability implementation note from the custom-header client guidance (per the proposed action here) and keep only the reject →tools/list→ retry guidance. Will open a PR stacked on #2972 shortly.fix(spec): replace Mcp-Param-* implementation note with rejection-ret…
fix(sep-automation): record staleness actions in the run summary