scavio-ai/hermes-agent

Scavio for Hermes Agent

Structured web data across 31 platforms plus an Extract endpoint for any URL, exposed as MCP tools and ready-to-use skills for Hermes Agent.

Built on the open Agent Skills standard, so the same skills also run in Claude, Cursor, Goose, OpenHands, and other compatible clients.

What you get

• A drop-in MCP server config connecting Hermes to mcp.scavio.dev
• 5 production-grade skills:
  • product-deep-research aggregate Amazon, Walmart, YouTube reviews, and Reddit threads into a buying brief
  • competitive-intel-brief Google, Reddit, and YouTube scan of a competitor
  • price-drop-watch Amazon vs Walmart side-by-side price diff
  • reddit-sentiment-pulse cross-subreddit sentiment aggregation
  • youtube-research-digest YouTube search plus transcript summarization

Coverage

31 platforms plus Extract, all behind one API key.

The hosted MCP server exposes 191 tools in total. The default surface registers 106 tools across 11 platforms, which is already a lot of context to carry in every session, so the config snippet in this repo narrows that further. See Tool surface below.

Install

The fastest path, once you have a Scavio API key:

# 1. Get a key (free: 50 credits on signup, one time, no credit card)
open "https://dashboard.scavio.dev/sign-up?utm_source=hermes"

# 2. Store it where Hermes can read it
mkdir -p ~/.hermes
echo "SCAVIO_API_KEY=sk_live_xxxxxxxxxxxx" >> ~/.hermes/.env
chmod 600 ~/.hermes/.env

# 3. Install this repo as a Hermes plugin
hermes plugins install scavio-ai/hermes-agent

Prefer to wire it up by hand, or want to tune which tools load? See INSTALL.md for the manual path, the config snippet, and troubleshooting.

Try it

Start a session:

hermes chat

Then ask:

Research the Sony WH-1000XM5 across Amazon, Walmart, YouTube reviews, and Reddit. Give me a buying brief with prices, review consensus, and red flags.

Hermes activates product-deep-research, fans out search across all four platforms via Scavio, and returns a structured markdown brief.

Tool surface

The Scavio MCP server exposes 191 tools. Loading all of them writes roughly 70k tokens into every session before you have typed anything, and exceeds the tool limit some clients enforce. Two knobs control this, and the snippet in config/hermes.config.snippet.yaml uses both:

• x-scavio-platforms on the MCP connection picks which platforms the server registers at all. It is additive against the keyword default, so default,zillow,sec means the default 11 platforms plus two more, and all means all 191 tools.
• tools.include in the Hermes server entry filters that down client-side to the exact tools you want in the picker.

The shipped snippet requests every platform these five skills actually call, then includes 18 tools. Widen either list when you want more of the catalog.

Pricing

Scavio is credit-metered. Cost varies by platform and by endpoint, so budget against the table in the docs rather than assuming a flat rate.

Verified examples: Google, Amazon, Walmart, eBay, Target, Airbnb, Zillow, Redfin, SEC EDGAR, UK Companies House and Meta Ads cost 1 credit; Threads, Yelp, Tripadvisor, Indeed, Capterra, Google Play and Home Depot cost 2; G2 costs 5; Kuaishou ranges from 1 to 40 depending on the endpoint. A few platforms price per endpoint rather than per platform, YouTube among them.

Plans:

• Free 50 credits on signup, one time, no credit card, no monthly refill
• Project $30/month 7,000 credits
• Bootstrap $100/month 28,000 credits
• Startup $250/month 85,000 credits
• Growth $500/month 200,000 credits

A single product-deep-research run typically makes 8 to 15 API calls. The credit total depends on which platforms those calls land on.

See plans

License

MIT. See LICENSE.

Support

• Issues and feature requests: GitHub issues
• Scavio API docs: scavio.dev/docs
• Hermes Agent: hermes-agent.nousresearch.com