LocalWebMCP

Can AI agents query structured local information through a standardized access layer?

A tool surface built around entity, geography and constraint fits agent behaviour better than a search box does.

Hypothesis — not a conclusion

If agents become significant consumers of local information, the access layer decides what they can read, how fresh it is and whether the source is attributed.

1. AI / Agent
   ↓
2. MCP access layer
   ↓
3. Local Graph
   ↓
4. Business Records
   ↓
5. Entities

• LocalWebMCP.com ↗

• —That agent queries are mostly scoped by place, category and constraint
• —That a small tool surface covers most local agent tasks
• —That attribution can be carried through every response

Access design turns out to be a product decision — what an agent may read, how often, and with what attribution shapes the whole layer.

Testing has not yet produced documented results. No public MCP service exists yet.

• Agents prefer general web sources over a specialized layer
• Standard schemas already cover the need
• Serving cost outweighs the retrieval benefit

• What is the right unit of response for an agent?
• How should rate, cost and attribution be handled together?

Draft a tool surface for entity lookup, geographic scoping and record retrieval, then write example agent transcripts against it.