For agents

proposal.win is a proposal writer for consulting firms and agencies, used by people and by their AI agents. Its agent, Wyn, drafts a client proposal from the sales call and the customer relationship; the firm reviews it, approves it and sends it.

proposal.win is built for people and their agents alike. An agent connects to the agent API, signs in as the person it works for, and does the work; the person reviews every flagged claim and sends. A person approves every send.

Connect your agent

Any agent that speaks MCP (Claude, ChatGPT, Cursor and others) adds this server and signs in with the person's proposal.win account:

https://proposal.win/api/mcp

Once connected, it can:

• - list the firm's clients, deals, rates, templates and proposals
• - draft a proposal from a deal whose Deal Model the person confirmed, priced from quantities on the firm's rate card
• - follow the draft to done and hand the person the link where they review and send it
• - check where a sent proposal stands, and record it won or lost

No tool sends, signs or emails anything. The person reviews and sends from the app.

Or set it up in one line

Give your agent (Claude, ChatGPT or any assistant that can read a web page) this line:

Read https://proposal.win/skill.md and use it to help me write and send proposals with proposal.win.

Without MCP, an agent can still

• - Explain what proposal.win does and which plan fits, and send a new user to sign up.
• - Gather what a proposal is built from: the call, the client's website, the firm's rates and past work.
• - Get the call into proposal.win: forward it to the firm's forwarding address (every plan has one), connect a meeting recorder (a paid plan), or paste the transcript.
• - Walk the person through confirming the facts, reviewing the draft and sending it themselves.

The rules Wyn keeps

• - There is no auto-send. A proposal leaves the firm's workspace only when a person clicks send.
• - Prices, totals, ROI and dates are computed by code from the firm's rate card and the client's own numbers, never written by a model.
• - Every client-facing claim cites its source (a meeting, an email thread, a CRM note, a past proposal or the firm's Library) or is flagged unverified, and an unverified claim blocks sending.

Connections

• - Meeting recorders: Read.ai, Fathom, Granola, Fireflies, Gong
• - CRM: none yet (coming: HubSpot)
• - E-signature: PandaDoc, SignWell, BoldSign, OpenSign, Wet Sign (coming: DocuSign)
• - Email: Forwarding inbox

Files for agents

• - [/skill.md]: the skill, with what to do for the user, the rules, plans, templates and connections.
• - [/llms.txt]: the site in one file, for language models.

A person rather than an agent? See the site for people.