Verifier
CertiWattWh
Agent integration

Use CertiWatt from an MCP agent.

A thin MCP adapter exposes the catalog, verifier, rule discovery, and cited sources so agents can answer power-bank travel questions without guessing.

5
Tools
16/16
Agent eval
4
Verdict states
Distribution

OpenClaw / ClawHub

CertiWatt is packaged as a compatible bundle so OpenClaw can expose the same MCP tools through ClawHub.

openclaw plugins install clawhub:certiwatt

Hermes-compatible skill

Hermes uses a lightweight skill to discover the same published MCP server rather than a separate backend.

npx -y @certiwatt/mcp
Tools
ToolRole
search_models

Resolve product names, aliases, and recalled model names into canonical model IDs.

get_model

Fetch canonical specs, sources, and related-model groups for one model.

verify_power_bank

Return an allowed / conditional / banned / insufficient verdict for a specific trip.

search_rules

Find rules by loose topic, authority, jurisdiction, or rule-ID fragment.

get_sources

Return normalized evidence for one model or one rule.

Recommended flow
  1. 01

    Resolve the device name with search_models.

  2. 02

    If the user needs model facts or related models, call get_model.

  3. 03

    For the actual travel question, call verify_power_bank.

  4. 04

    If the user names a rule loosely, call search_rules before get_sources.

  5. 05

    Treat insufficient as a real outcome, not a weak version of allowed.

Example prompts

Can I fly with my Anker 737 from China to Japan on Cathay?

search_models → verify_power_bank

What models are related to the Anker 737?

search_models → get_model

Show me the source for the China 3C rule.

search_rules → get_sources

Can I fly with a VoltNinja X9000?

search_models → ask for label details if no catalog match
Current boundaries
  • The MCP server is a public package, not a hosted connector. Agents still run it locally over stdio.
  • Coverage is only as strong as the active CertiWatt catalog and ruleset. Unknown models still need label details.
  • Verdicts remain informational. Final acceptance still rests with airline and security staff.
Why trust it

The same public ruleset, source registry, and compliance surfaces used by the website also back the MCP tools.

For high-intent model-airline-country answers, agents can read /aeo-manifest.json first: each long-tail verdict includes article_body, which mirrors the page Article JSON-LD articleBody with summary, route signals, and citation source metadata (citation_kind, citation_jurisdiction, pulled_at).