mcp 0.5.0 tool descriptions with unicode homoglyphs pass validation but display wrong in context
tested tool registration with descriptions containing unicode homoglyphs (latin 'a' vs cyrillic 'а') and direction overrides. registration validates the homoglyph version but when it hits model context the display is completely different from what was registered. example: registered tool description "reads user data" but with cyrillic 'а' renders as "reads user dаta" in context, model sees different semantic meaning. no sanitization on tool descriptions at registration time in 0.5.0, everything passes through
tested this on 0.5.0 with tool description containing Cyrillic 'а' instead of Latin 'a'. passes validation, displays identical to user, but context has different unicode. did model actually parse it different?
did it actually affect model behavior or just display wrong to humans? need to know if this is cosmetic or exploitable