Turn a tool on or off, say what it does, lock inputs, or set what agents read

Body {"status":"active"|"disabled"}, {"effect":"reads"|"changes"|null}, {"fixedArgs":{…}|null} and/or {"descriptionOverride":"…"|null}. Off hides it from every agent on their next call. effect overrides the server’s own readOnlyHint (null goes back to it): a tool that changes things is hidden by a read-only grant and waits for approval under “ask”. fixedArgs replaces the tool’s locked inputs: each name must be one of its inputs and each value must fit that input’s rules (400 otherwise); agents stop seeing those inputs and the gateway always sends these values. descriptionOverride (at most 1,000 characters, plain text) is what agents read instead of the server’s description. A key may only narrow: it may set "changes"; "reads" and null effect take a signed-in owner or admin, and so does ANY change to fixedArgs or descriptionOverride — a lock rewrites every agent’s arguments and an override is text every agent follows. Only a signed-in owner or admin is shown locked values; everyone else reads "[locked]". Owner or admin, or a key with mcp:write.