Reddit asks how to defend MCP tools against post-approval definition changes
Agile_Wedding9018 · reddit · 2026-07-25
A Reddit thread asks how people handle MCP tools whose definitions change after approval, raising concerns around MCPoison-style attacks and tool-output prompt injection.
The poster asks whether production users:
- re-hash or re-check tool definitions on every call,
- sandbox tool execution,
- treat tool output as untrusted,
- or simply trust the MCP server and hope for the best.
The practical concern is that a tool can be approved once with a harmless description, then later be updated to quietly do something else, while the model continues trusting it.
More from Safety
- A 13KB Markdown file can hijack Claude Code and Codex via your Downloads folder — pbaylies · 2026-10-26
- If an agent can rewrite itself, does your consent still mean anything? — tallmetommy · 2026-10-26
- 109 of 529 AI Agent Incidents Involved No Adversary, Arguing Red Teams Must Test Normal Operation — anacondainc · 2026-10-26
- Cryptographer Matthew Green: if your code looks vulnerable, fix it now — matthew_d_green · 2026-10-26
- Filtering secrets on the write path: why agent memory deserves more suspicion at ingest than retrieval — Classic-Being-7960 · 2026-10-26
- After OpenCode key leak: 5 practical rules for keeping keys away from agents — jacobholm_dev · 2026-10-26