Your evidence, available to your assistant

Evident runs a remote MCP server, so an AI assistant can read your reviews, catalogue, orders and loyalty programme — and, where you allow it, act on them. No plugin to install and no client to write.

https://mcp.evidentugc.com/mcp
Read the setup guide

Questions you can just ask

The analytics tools return real aggregates rather than a sample, so an assistant answering with numbers is answering with your numbers. Most of the value is in reading — pulling together things that would otherwise mean opening five screens.

Full tool list and setup in the agent access documentation.

Which products dropped below 4 stars this quarter, and what are the negative reviews complaining about?

Find reviews below 3 stars from the last month with no reply, and draft a reply for each.

Read our FAQs and the last 200 reviews for this product. What are customers repeatedly asking that we haven't answered?

Summarise the pending moderation queue and flag anything with personal data or a safety concern.

19 read tools

Catalogue, orders, customer activity, reviews, review and evidence analytics, FAQs, galleries, loyalty accounts, transactions, rewards and redemptions.

9 write tools

Moderate a review or a whole queue, reply, send a review request, moderate a gallery item, create and update FAQs, adjust points, create a reward.

Scoped by key

An agent can do exactly what its API key can do. Hand over a read-only key and it stays read-only.

Read-only annotated

Read tools are marked as such, so a well-behaved client can show you which calls actually change something.

Tenant-scoped

Every call names a store. There is no cross-store query and no "all stores" mode.

Nothing to install

A remote server over streamable HTTP. Paste a URL and a key.

Worth knowing before you connect one

Review content is written by the public. An assistant reading reviews is reading text a stranger wrote. Have it draft replies rather than post them, and do not build unattended automation that acts on review bodies.

Adjusting loyalty points mints value. Points redeem into store credit and discount codes. Leave that scope off the key unless you specifically want it.

Bearer token is the connection method today. Signing in with your Evident account instead of pasting a key is built but not yet enabled.

Common questions

What is MCP, in one paragraph? +

The Model Context Protocol is a standard way for an AI assistant to call tools on a remote server. Evident runs one at a fixed URL, so instead of writing an integration you paste that URL and an API key into Claude, ChatGPT, n8n, Zapier or Make, and the assistant discovers what it can do. There is nothing to install and nothing to host.

What can an agent actually do? +

Twenty-eight tools: nineteen that read and nine that write. The reads cover your catalogue, orders, customer activity, reviews, review and evidence analytics, FAQs, galleries, and loyalty accounts, transactions, rewards and redemptions. The writes moderate a review or a whole queue, publish a reply, send a review request, moderate a gallery item, create and update FAQs, adjust a points balance and create a reward.

How do I stop an agent doing something I did not intend? +

By what you put on the key. An agent can do exactly what its API key can do, and a key can be narrowed to read-only or to specific resources, so a read-only key stays read-only no matter what the assistant is asked. Read tools are also annotated as read-only in the protocol, which lets a well-behaved client show you which calls change something and which do not.

Can it see more than one store? +

No. Every call names a store environment, and there is no cross-store query and no all-stores mode. An agency or a multi-store merchant hands over one key per store rather than one key that sees everything.

Do the analytics tools return a sample? +

No, they return real aggregates, which is the reason they are worth asking rather than exporting. An assistant answering "which products dropped below four stars this quarter" is answering from your whole corpus, not a page of it.

Is it safe to let an assistant read reviews? +

Reading is fine; acting unattended on what it reads is not. Review bodies are written by the public, so an assistant reading them is reading text a stranger composed, and a stranger who wants to can compose text aimed at your automation. Have it draft replies for you to approve rather than post them, and do not build unattended workflows that act on review content.

How does it authenticate? +

A bearer API key today. Signing in with your Evident account instead of pasting a key is built but not yet enabled, so until it is, create a key with the narrowest scope that does the job and treat it like a password.

Can I use it without an AI assistant at all? +

Yes. The same surface is available as a normal HTTP API, and outbound webhooks will tell you when something happens rather than making you poll for it. MCP is a convenience for assistants, not the only way in.

Everything else about agent access

This page is the overview. The detail is written up in full, and you do not need an account to read any of it.

Where a setting lives and what it does.

How to run it, not how to configure it.

The third-party tools it connects to.

  • Claude

    Sign in with your Evident account; no key to paste.

  • ChatGPT

    Connect with an API key as a bearer credential.

  • n8n

    Evident as an MCP tool inside your workflows.

  • Zapier

    Triggers and actions across 8,000+ apps.

  • Make

    The same tool surface in a Make scenario.

Connect your assistant

Create a read-only key and paste one URL. Included in every plan.