Complete vehicle record
Shows a completed call whose business goal is fully resolved.
I am from KNWN Demo Motors. Decode VIN 1HGCM82633A004352 for a 2003 vehicle and tell me whether the record is ready for a parts quote.
Vehicle quote MCP + analytics
A real NHTSA workflow and the usage evidence KNWN ships with every MCP build.
Four real NHTSA workflows. No key, canned result, or seeded analytics.
Use this remote URL as a custom connector in Claude.
After any tool call, ask Claude to inspect the returned analytics reference.
Show me what this MCP recorded from that request, including any friction it detected.
Each button calls the same instrumented tools exposed through the MCP.
Shows a completed call whose business goal is fully resolved.
I am from KNWN Demo Motors. Decode VIN 1HGCM82633A004352 for a 2003 vehicle and tell me whether the record is ready for a parts quote.
Shows a correctly completed call that identifies why the quote cannot continue.
I am from KNWN Demo Motors. A customer sent VIN 5UXWX7C5*BA for a 2011 vehicle. Decode it for a parts quote and tell me exactly what is missing.
Shows a search tool returning a real result set and result count.
I am from KNWN Demo Motors. List the Toyota models available in NHTSA data for model year 2024.
Runs the same request twice in one activity and flags the retry as observable friction.
I am from KNWN Demo Motors. List the Rivian models in NHTSA data for model year 2024, then retry the same request.
Live records from the four controlled web examples above.
Completion asks whether the tool returned normally. Resolution asks whether the business goal was fully fulfilled. A partial result can be a correctly completed call. Rates exclude calls still in progress.
The evidence layer records the workflow, not the model's hidden reasoning.
| Time | Tool | Query sent to tool | Company | Execution | Result | Time | Friction |
|---|---|---|---|---|---|---|---|
| No web-demo calls recorded yet. Run an example above. | |||||||
The exact argument received by the MCP. It is marked agent-transcribed unless a trusted host supplies the original prompt.
Verified only through validated authentication or tenant mapping. The public demo is self-declared and marked asserted.
Tool execution time and observed activity span. It is not the duration of a Claude or ChatGPT conversation.
Observable friction such as missing inputs, empty results, retries, timeouts, auth failures, or rate limits.