Proofread your website from your AI tools
Verant runs a Model Context Protocol (MCP) server, so any agent that speaks MCP — Claude, Cursor, and the rest — can proofread a page or crawl a whole site without leaving the tool you already work in.
Last reviewed: 2026-06-21
Verant exposes a Model Context Protocol (MCP) server at verant.ai/mcp, listed in the official MCP registry as ai.verant/verant. Connect it once over OAuth from any MCP client — Claude, Cursor, or your own agent — and the agent gets three tools: proofread_url for a single page, proofread_site for a full crawl, and get_scan for a scan's status and results.
Scans run on your Verant account and plan, the same engine the web app uses, and every correction is still refuted by a second, different-vendor model before it reaches your agent. There is no anonymous or trial access over MCP — the connection is authorized to you, and your quota is checked on every call.
What the MCP server does
The MCP server is the same Verant engine, reachable from an agent instead of the dashboard. proofread_url loads one published page, renders it, and proofreads the visible copy. proofread_site crawls a public site and proofreads every page it finds, up to your plan's ceiling. get_scan returns the status and the verified corrections for a scan your account owns, so a long crawl can be started and then polled.
Each correction quotes your exact text, nothing is auto-applied, and a second different-vendor model has to fail to refute a suggestion before it is returned. The agent receives a short, verified list — not a wall of raw suggestions it has to sort.
Why proofread from your agent
You are already drafting, editing, and shipping copy inside Claude or Cursor. With Verant connected, you can ask the agent to check the page you just changed, or crawl the whole site before a release, and get the answer in the same thread — no tab-switch, no paste, no separate dashboard trip.
Because the result is a verified list rather than a pile of maybes, an agent can act on it directly: surface the real issues, draft the fixes for your review, and move on. The trust pass that makes Verant useful to a person is exactly what makes it safe to hand an agent.
Built for agents, safely
Every MCP call is authorized over OAuth and acts as your Verant user — there is no shared key and no anonymous tier. Your plan and remaining quota are resolved from your account on every call, so an agent can never scan beyond what you pay for, and one account can never see another's scans.
The verbatim rule still holds over MCP: Verant flags your exact text and never edits your site. What the agent gets back is a list of verified corrections to act on, not changes already made.
Connect Verant to your AI client
- 1
Add Verant's MCP server in your client and point it at https://verant.ai/mcp (Streamable HTTP). It is listed in the official MCP registry as ai.verant/verant.
- 2
Authorize over OAuth — sign in to your Verant account when the client prompts. The connection acts as you, on your plan.
- 3
Ask your agent to proofread — call proofread_url for one page or proofread_site for a full crawl, then get_scan to read the verified results.
How verification works
Most proofreading agents show you every suggestion and make you sort the good from the bad. Verant runs an adversarial second pass — Claude Sonnet proofreads, then GPT-5 tries to break each correction. What survives is what we show you. Verbatim is sacred: every flag quotes your exact text; we never auto-apply fixes.
Explore Verant
Related reading: the full website proofreading pass, automatic weekly monitoring, and transparent, self-serve pricing.
Frequently asked questions
Which AI tools can use Verant?
Any client that supports the Model Context Protocol — Claude, Cursor, and other MCP-capable agents. Verant's server is Streamable HTTP at verant.ai/mcp and is listed in the official MCP registry as ai.verant/verant.
Do I need a paid plan to use the MCP server?
Yes. MCP scans run on your Verant account and plan, and there is no anonymous or trial tier over MCP. The connection is authorized to you over OAuth, and your remaining quota is checked on every call.
What can the agent actually do?
Three tools: proofread_url checks a single published page, proofread_site crawls and proofreads a whole public site up to your plan's ceiling, and get_scan returns the status and verified results of a scan your account owns.
Are corrections from the MCP server verified too?
Yes — the MCP server runs the same engine as the web app, so every correction is still refuted by a second, different-vendor model before it is returned. Your agent receives a short verified list, with your exact text quoted, not raw suggestions.
Does the agent change my site?
No. Verant flags your exact text and never edits or auto-applies anything. The agent gets back a list of verified corrections to act on; any change is yours to make.