Cartograph reduces AI agent tool discovery from O(n) to O(k)
Cartograph is a federated Model Context Protocol (MCP) proxy that changes how AI agents find and call tools. Instead of traversing every catalog entry, the system exposes only a few proxy tools, lowering the search complexity from O(n) to O(k). The authors claim the approach is more efficient for large, distributed tool collections.
Key points
- Cartograph reduces tool discovery complexity from O(n) to O(k)
- It exposes only three proxy tools in a 22‑server, 374‑tool deployment
- Recall@5 improves from 0.592 to 0.816 with only 475 tokens used
The proxy uses three mechanisms. First, operator‑attested capability cards are Ed25519‑signed descriptions generated by the deploying operator, not by the publisher. Second, Rift performs a three‑layer confusable‑cluster analysis—density clustering, query‑margin analysis, and token diagnosis—to group similar tools. Third, a two‑stage retrieval ranks servers before ranking individual tools. In a test with 22 servers and 374 tools, Cartograph exposed only three proxy tools.
Performance results show a recall@5 of 0.816 versus 0.592 for a Jaccard keyword baseline. The top‑5 discovery exchange uses 475 tokens instead of 42,450 tokens under full‑catalog accounting. Rift identified 49 confusable clusters, including four high‑risk clusters. Gateway measurements over ten trials added only 5 ms mean latency (0.8 %) compared to direct stdio MCP calls.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Researchers propose attention-free model mixing with autoencoders for masked language tasks · 1 src
- Study identifies under 1% of BERT neurons driving AI-Text detection · 1 src
- Researchers release Atelier for cryoEM map analysis via hypernetworks · 1 src
- Researchers find intuitive prompts improve LLM social media simulation · 1 src
- Researchers introduce Benchy, a standardized language for AI task benchmarks · 1 src
Comments
via GitHub Discussions