This path is for people with technical skills, or with an AI agent of their own: Claude Code, Claude Desktop, claude.ai, Cursor and 70+ other skills-capable agents. No agent? Use the iOS or Android app instead.
1. Install the skill
One command copies the skill into your agent’s skills directory:
npx skills add commonspecs/skills Alternatively, copy the skill from the commonspecs/skills repository straight into your agent’s skills directory — ~/.claude/skills/ for Claude Code.
2. Ask
Ask your agent about a product or a service and it picks the right call:
- ask for advice on what to buy in a category — “recommend me a good toothpaste” — and the agent goes through that category’s ranking with
search_products - add a condition on a parameter (“only linen”, “heavily carbonated”) and it narrows the category with spec filters, matched against the recorded facts
- ask about one spec of a product you name — the agent finds it by brand and model with
search_products, then pulls the details withget_product - ask it to compare product X with product Y — the agent uses
compare_products - take a photo of a label — the agent reads the values and the EAN, then looks them up; the photo never leaves your device
Recommended: add the MCP connection
We recommend pairing the skill with the MCP connection — it talks to your account over OAuth, so there is no token to create or manage. Clients with native connectors (Claude Desktop, claude.ai, Cursor, …) sign in in the browser, and the skill detects the connection and uses its tools. A client that cannot run skills can use MCP on its own.
The server endpoint:
https://api.commonspecs.com/mcp Claude Code:
claude mcp add --transport http commonspecs https://api.commonspecs.com/mcp Other clients (Cursor, VS Code, Claude Desktop connectors) take an HTTP MCP server config with the same URL — OAuth-capable clients sign in on first use. For a stdio-only client, bridge it:
npx mcp-remote https://api.commonspecs.com/mcp Don’t want MCP? Put an API token generated on your profile into the environment instead — the skill falls back to direct calls with it:
export COMMONSPECS_API_TOKEN="cs_live_…" Your buying preferences live on your account and ride along on every read — the server scopes results to your market and locality, and passes your quality goal and contribution mode to the agent to apply.
What it does with the answer
The skill reports every fact with its confidence, treats weakly-corroborated values with caution, shows both sides of a disputed value, and surfaces what a record is still missing (missing_fields). It never invents a value it didn’t get back. When a field would benefit from another source and you’re interested, it can fetch one and send it back — the conversation itself feeds the database.
How the ranking is built — see the methodology.