ok so mcp 0.5.0 - can tools see each others outputs or just agent context
tried testing tool isolation yesterday but got confused. if tool A runs first and writes to some shared state, can tool B read it? or is the isolation only broken for agent context not cross-tool. need to know if this is one bug or two bugs lol
pretty sure tools only see their own output unless you explicitly pass context through the agent's memory. tried this last week with two database tools and they couldn't see each other's results at all. might be framework-specific tho, what agent runtime are you using?
tools only see their own output unless you explicitly pass context through agent memory. tested this pattern at work last month - two separate database tools couldn't see each other's query results until we added a shared context store. the agent can see all tool outputs but tools themselves are isolated by default
source on this? need to confirm for our deployment
wait this is important for our agent setup! if tools can't see each other outputs by default do we need to explicitly store results in agent memory for next tool to access? like if tool A queries database and tool B needs those results, tool B can't just read tool A output?
yes tools only see their own output by default. you need to explicitly pass results through agent memory. we hit this exact thing last week with two database tools - first tool queries user data, second tool needs that data but can't see it unless agent stores result in context