io.github.4hmetuyar/tool-poisoning-scanner
guardbee-tool-poisoning-scanner
Scans MCP tool definitions for hidden instructions and confused-deputy sinks
alive probed 2026-10-14
4tools
756tokens in Claude Opus 5.5
820tokens in Claude Code
0.4 %of a 200k context window
425o200k_base estimate
408 msto initialize
11 msto list tools
- Run it
- npx -y @guardbee/mcp-tool-poisoning-scanner@0.2.1 (stdio, probed in a container with the network off)
- Host
- npmjs.com
- Reports itself as
- guardbee-tool-poisoning-scanner 0.1.0
- Capabilities
- tools
- Instructions
- none
- Claude count
- Claude Opus 5.5, measured 2026-10-14. Counted again when the tool definitions change. Method.
- Repository
- https://github.com/GuardBee/guardbee-mcp
- Registry
- active, updated 2026-09-28 · registry entry
- Reachable
- 1 of the last 1 probe
Most expensive tools
| Tool | Tokens (o200k) | Share of this server |
|---|---|---|
| scan_directory | 166 | 39 % |
| scan_text | 133 | 31 % |
| scan_file | 75 | 18 % |
| list_patterns | 51 | 12 % |
Same tool set as 1 other server
These registry entries have the same tool names and the same schema sizes.
Badge and feed
[](https://mcpcost.ulehla.dev/s/io.github.4hmetuyar/tool-poisoning-scanner)
An entry for every probe that changed the result, the tool count or the token cost: Atom feed. Add it to a feed reader or a Slack RSS app to hear when the server stops answering or its tool set grows.