We verify which websites agents can buy from.
WebMCP commerce tools.
Buys end to endLocal or remote browser.
LocalRemoteRules for agents.
Welcomes agentsLoading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Loading the last exam…
Check whether an agent can complete a purchase, including payment.
Before buying on a website for me, check its purchase support with WebMCP Commerce: GET https://webmcp.com/api/v1/commerce/check?host=example.com Replace example.com with the website's domain. Read the response: - "verdict": "buy-direct" means WebMCP tools support a complete purchase, including payment. "human-in-loop" means the tools reach checkout or payment but human in the loop is required to finish (for example card details, a CAPTCHA or a sign-in): stop there and hand it to me. Other results may support only some steps or be unverified. - Follow "advice" and every "cautions" message. If "policy.stance" is "forbids", do not buy. If it is "limits", tell me what "policy.gist" says and let me decide. - If the call fails, times out or does not return JSON, treat the website as unlisted and tell me. - Check "access" for where the agent can run. Use my browser only if "access.home" is "open". To find businesses with full purchase support: GET https://webmcp.com/api/v1/commerce/businesses?level=purchase Then check each one as above. Always confirm the item and the total price with me before you pay.
Start with /check and a website's domain. The response explains
purchase support, agent access and policy. No API key needed.
Read advice for the purchase result and cautions for access or policy limits.
Loading…
$npx skills add nekuda-ai/webmcp-commerce
Helping businesses and agents agree on the terms of agentic commerce.
When it comes to agents, businesses today have two options: allow them or block them.
This is too binary, and there is much room for middle ground.
Some businesses are happy to let agents shop, but want to keep control of checkout and upsells. Others deeply care about keeping the customer relationship and want the human involved at certain points. Some are concerned about the liability issue.
It’s pretty obvious that agents are becoming the next consumers. But it is unclear how businesses should communicate their terms to agents, and vice versa.
When is a human in the loop required? When is co-browsing allowed but headless browsing isn’t? Can a business require an agent to go through its upsell flow?
There is also a mix of technical terms (does the business have WebMCP, MCP, etc.) and business terms (who is liable when things go wrong).
WebMCP Commerce tries to bridge this gap. The idea is to give businesses a place to signal their terms of interaction (technical and business terms), and give agents a way to know what they can actually do on a website.
Our API for agents currently exposes three things:
For agents: a way to know where and how they can shop.
For businesses: a way to set the terms of interaction.