Every AI product has the same uncomfortable line on its P&L: inference. The model calls that make the product useful are also its biggest variable cost, and they scale with usage — the more people love it, the more it costs to run. Bring Your Own Model (BYOM) flips that line to zero by letting users run the AI on the subscription they already pay for. Here’s what BYOM is, why the Model Context Protocol (MCP) makes it possible, and what it does to pricing.
Classic SaaS has near-zero marginal cost: one more user is one more row in a database. AI SaaS doesn’t. Each meaningful action — summarize, rewrite, classify, generate — is a model call with a real per-token price. Margins compress exactly when a product succeeds, which forces credit caps, rate limits, and “fair use” language that users feel as friction.
The first mitigation was Bring Your Own Key: the user connects their own API key, and the product runs inference on it. The provider’s COGS for that user drops to near zero — but the user still pays per token, and usage is metered against their key. It shifts the cost; it doesn’t remove it. And it requires the user to manage a key and watch a balance.
Bring Your Own Model removes the per-token cost entirely. Millions of people already pay a flat monthly fee for an AI agent — Claude Code, for example. That subscription is sunk cost; the marginal price of one more generation inside it is effectively nothing. BYOM lets a product tap that: the user’s own agent does the reasoning, and the product just supplies data, structure, and actions. No API key, no token meter, no per-use charge — for the user or the vendor.
The Model Context Protocol is the missing piece. With MCP, a product publishes two kinds of capability:
The user’s agent becomes the orchestrator: it calls the prompts (on its model) and the tools (on the product), stitching them into a workflow. The expensive thinking happens on the user’s side; the product keeps the parts that are genuinely its moat. That division is what makes the COGS line go to zero.
When inference COGS is ~$0, you can do things metered AI products can’t:
The remaining costs are the non-AI ones — scraping, rendering, storage, automation — which are small and bounded. You price and cap around those, not around tokens.
ResumeAlign applies this directly. Connect Claude Code over MCP and your agent tailors each résumé to each job on its model; ResumeAlign scrapes the posting, renders the ATS-clean PDF, and drives the application. The tailoring — the expensive part — costs nothing, because it runs on your Claude Code subscription. That’s why the MCP Plan can offer unlimited auto-apply at $19.99/mo, and the free tier can include real monthly volume. Same product, structurally cheaper, because of who owns the model.
Want the how-to? Read Auto-Apply to Jobs with Claude Code and how to schedule it.
What is the difference between BYOK and BYOM?
BYOK (Bring Your Own Key) runs the product on your API key — you still pay the model provider per token. BYOM (Bring Your Own Model) runs the reasoning inside your own agent (e.g. Claude Code), which you already pay a flat fee for — so there’s no per-token cost to you or the vendor. BYOK shifts the cost; BYOM removes it.
How does MCP enable BYOM?
The Model Context Protocol lets a product expose tools (the things only it can do — scrape, render, take actions) and prompts (the reasoning, run on the user’s model). The user’s agent orchestrates both: thinking happens on the user’s model, actions on the product. That split is what drops the product’s inference COGS to near zero.
Why is a BYOM free tier hard to abuse?
A free tier that runs on the vendor’s model can be milked as “free Claude.” A BYOM tier can’t — the user supplies the model, so there’s no vendor inference to abuse. The only costs left are non-AI (scraping, rendering, storage), which are small and easy to cap.