Your buyer opened the proposal twice yesterday and spent most of their time on the pricing section. Qwilr recorded all of it. Your CRM knows none of it, and neither does the rep who owns the deal, until somebody opens the analytics dashboard and types it into the deal record.
Model Context Protocol, or MCP, is what closes that gap. It lets the AI assistant your team already uses reach into Qwilr, read what buyers are doing, and act on your pages.
Most of the attention on AI in sales has gone to drafting documents faster. We think the part that decides deals is what happens after you hit send. That is why our connection does more reading than writing, and why it cannot touch the content of a page once it exists.
What is MCP?
MCP is a shared standard for connecting AI assistants to the tools you already use. Instead of a custom integration between every assistant and every app, each app publishes one connection that any assistant can use. Qwilr publishes one.
Connect Qwilr to an assistant that supports MCP and it works with your live proposal data, rather than whatever someone remembered to paste into a prompt.
What an assistant can do through Qwilr's MCP
An assistant can read what is happening across your proposals and act on your pages. Here is the whole picture, including what it deliberately cannot do.
| Action | Available | What to know |
|---|---|---|
Search pages by client, status or other criteria | Yes | Across the whole account, within your own permissions |
Read engagement analytics | Yes | Block-by-block view durations. Viewer email, name, IP and precise location are stripped; sessions carry device, browser and country only |
Identify named viewers | Yes | A separate call, returning name, email and whether the identity is verified. Anonymous views come back empty |
Report pipeline value and close rate | Yes | Admins can report on any owner. Everyone else sees only their own pages |
Create a page from a template | Yes | Uses one document automation credit. There is no de-duplication, so running it twice makes two pages |
Turn an existing page into a template | Yes | Copies the content but not the variables. You add those in the editor, once |
Edit the content of an existing page | No | The connection exposes no way to do it |
Take a page live, or unpublish it | Yes | Freely reversible |
Mark a page accepted | With confirmation | Locks the page permanently. It cannot be changed afterwards |
Mark a page declined | With confirmation | Records the deal as lost |
Send a page to a client | No | Nothing leaves Qwilr on its own |
What an AI assistant can and cannot do in Qwilr over the MCP connection.
Everything runs under the Qwilr login you connect with, so whatever that person can see and do in Qwilr directly is the boundary the assistant works within. The same security posture applies as anywhere else in the platform.
What that looks like on a real deal
Take a six-person team pitching 30 sponsors for a conference. They have one sponsorship deck built as a Qwilr template, with variable fields for each sponsor's name and package tier.
Building that template is the part a person does, once. If the team has only been duplicating a page by hand, the assistant can turn that page into a template first, then someone adds the variables in the editor.
After that, 30 personalised pages is one request instead of an afternoon. Pages created without a folder land at the top level, so say which folder you want. And asking twice will give you 60 pages, because nothing is de-duplicated.
Once they are live, the question changes from who has been sent a deck to who is reading one. The sales lead asks which sponsors have opened theirs and where they spent time, and gets back a ranked answer: three reviewed the pricing tier twice, several skimmed and left, the rest have not opened at all. Nobody opened a dashboard. It is the same proposal analytics the product has always had, reached by asking instead of clicking, and it comes back with named viewers wherever the viewer identified themselves.
Across 1,064,667 proposals we analysed in 2025, the acceptance rate nearly doubled when at least two additional people viewed a proposal within the first five days. Knowing that a second and third person opened the deck on Tuesday is worth more than shaving ten minutes off building it.
Using Qwilr alongside your other tools
If your assistant is connected to several tools, it can chain them. A proposal view in Qwilr combines with account detail from your CRM to decide who needs a follow-up, then drafts that follow-up in your email tool for a rep to edit and send.
Some combinations to test, depending on what you have connected:
- Pair proposal engagement with CRM context to rank which deals deserve attention this week
- Draft a follow-up in your email or messaging tool off the back of a proposal signal
- Turn a signal into a next step elsewhere: a task, a calendar entry, a line in a reporting sheet
These are workflows your assistant assembles, not features Qwilr ships. Qwilr's server exposes proposal data and page actions and nothing beyond that, and the rep still sends the email.
Separately, and through a different mechanism, Qwilr has native integrations with HubSpot, Salesforce, Microsoft Dynamics, Zoho and Pipedrive. That is the route to take if your CRM has no MCP server of its own.
Once you are connected, seven prompts worth trying first covers the ones we reach for most, from checking which deals need attention to bulk-archiving old pages.
How to connect it
You need three things: an assistant that supports remote MCP servers, a Qwilr account to sign in with, and a template if you want to generate pages from one.
Add https://mcp.qwilr.com in your AI tool's connector settings. There is no API key; signing in with your standard Qwilr login handles access, and the connection inherits that login's role and permissions.
Access to the MCP is included on every plan, including legacy Business and Enterprise, and reading your proposal data costs nothing. Creating pages from a template runs on the newer Starter, Growth and Scale plans and draws on document automation credits.
Frequently asked questions
Access is included on every plan, including legacy Business and Enterprise, and reading your proposal data is free: who viewed a page, how long they spent, what is in your pipeline. Creating pages from a template runs on the newer Starter, Growth and Scale plans and draws on document automation credits. How far back analytics reach depends on your plan, between one and five years, and a request for a wider window is quietly narrowed to whatever your plan allows.
The page locks and cannot be changed afterwards, which is why accepting requires you to confirm first rather than happening on a single instruction. Declining works the same way and records the deal as lost. Taking a page live or pulling it back to draft needs no confirmation, because both are reversible. Pages with an Accept or Quote block can only be accepted by the recipient through the block, not by the assistant.
No, and not as a matter of policy: the connection exposes no tool for it. An assistant can create a page from a template and fill its variables at the moment of creation, manage status, and archive, delete or restore pages. Changing the words on a page that already exists has to happen in the editor.
Our MCP uses all the same security features that you've come to know and appreciate with using the Qwilr Platform. All your data is encrypted (in transit and at rest) and you're granted the same permissions you would if you were logging into the platform. We will only follow the instructions that you provide to us when interacting with your account and data.
Engagement analytics come back with viewer email, name, IP address and precise location removed, so a question about how a proposal is performing returns device, browser and country only. Named identity is a separate, deliberate call, and it returns a name and email only where the viewer identified themselves. Anonymous views come back empty.
