Why We Stopped Metering MCP Tool Calls

When an AI assistant has a security tool, it should use it constantly. Every package it suggests. Every CVE someone pastes into the chat. Every "is this safe to merge?"

Metering works against that. If every tool call costs something, the people setting up the agent start rationing. Fewer checks per session. Caps per user. Somebody in finance asks why the bill spiked during a busy sprint. The checks that get cut are the boring, frequent ones, which are exactly the ones that catch things.

So MCP tool calls on Safeguard are now unlimited on every plan. Free included. No monthly cap, no per minute peak, no charge.

What that looks like in practice

Connect Safeguard to Claude, ChatGPT, Cursor or your own agent. Then let it check whatever it wants:

• look up a package before suggesting it
• pull the CVEs for a dependency and whether any are known exploited
• check whether a finding is reachable in your code
• look up a fix version

None of that counts against anything.

What still costs money

Being straight about the edges:

• The API is pay as you go. The first 10,000 deterministic calls each month are included. After that, scans and lookups are $0.10 per 1,000 calls.
• Model usage is billed at the provider's list price. We don't add a markup.
• Plans still differ on features. Details are on the pricing page.

Why now

More of the security questions engineers ask now go through an assistant first. If the assistant can't afford to ask Safeguard, it guesses. We'd rather it asked.

See how to connect Safeguard to your assistant.

Try it yourself

Click through it at your own pace, or open it full screen.

Connect Safeguard to your assistant, then try three questions in one sitting: ask it to check a package before you install it, to list the known exploited CVEs in one of your projects, and to suggest a fixed version. Count how many tool calls that took. Then remember none of them counted against anything. More: use Safeguard from Claude.

Quick check