BLog
/
/
Rate Limits: Keep One Runaway Agent from Spending Everyone's Month
Content
↳
Example H2
↳
Example H3
↳
Example H4
↳
Example H5

Your task allowance sets how much your account can do in a month. Rate limits decide how fast any one person, team, AI client or tool is allowed to spend it.

It usually happens on a quiet afternoon. Someone on the operations team asks their AI assistant to "check every open sales order against inventory and tell me which ones will be late." The assistant starts calling the "Get data" tool on your PopdockAI MCP server. It reads one page of orders, then the next, then checks stock for each line, one call at a time. Nobody is watching. By the time the person comes back from a meeting, the assistant has made a few thousand tool calls, and a good share of the account's monthly tasks went with them.

Nothing broke, exactly. Every call was allowed and every answer was correct. The trouble is that one curious question just used up capacity the rest of the company was counting on, and you only find out when the month runs short.

Rate limits are how you stop that. They let an administrator say, in plain terms, "no more than this many calls per minute, hour or day" for a user, a team, an AI client, a single tool, or every user by default. This article explains the problem they solve, how they work, and when to reach for them.

The Problem with a Monthly Total Alone

PopdockAI already meters usage. Every MCP tool call spends a task from the account's monthly allowance, and when the allowance is gone, it is gone until the month resets. That protects your bill. It does not protect your colleagues.

A monthly total can't tell a burst from a busy month. Ten thousand calls spread across thirty days by a hundred people is healthy use. Ten thousand calls from one agent in an afternoon is a loop. The allowance counts both the same way, and only notices once the total runs out.

AI agents don't get tired. A person clicking through a list makes a few requests a minute. An agent working through a task can make dozens a minute and keep going for as long as the task lasts. A badly worded prompt, or a script that retries on every error, will keep calling until something stops it.

The source system pays too. Every call to a list goes through a connector to your ERP, CRM or finance system. A burst of calls against Business Central or NetSuite is load on that system, and on everyone else using it at the same moment.

What a Rate Limit Does

A rate limit is a rule with four parts: what it applies to, which calls it counts, a period, and a maximum.

  • Who or what it applies to. A user, a team, an MCP client, a single MCP tool, or the account's default for every user.
  • Which calls it counts. MCP tool calls from AI clients, or API calls to the Popdock API, including calls from the command line tools. MCP rules can cover every tool, every tool on one server, or one tool.
  • The period. Per minute, per hour or per day.
  • The maximum. The number of calls allowed in that period.

When a call arrives, PopdockAI checks it against every rule that matches. If any one of them is used up for the current period, the call is refused, and the refusal says whose limit it was and how long to wait: "Rate limit reached. Your team is limited to 500 MCP tool calls per hour, shared between its members. Try again in 12 minutes." The AI client receives that sentence as the tool's error, so it can tell the person in the conversation why it stopped. An API caller gets the same sentence with an HTTP 429 status and a Retry-After header, so a script can back off for exactly as long as it needs to.

The difference is easy to say:

  • Without rate limits: an agent in a loop runs until the task finishes or the month's allowance runs out, whichever comes first.
  • With rate limits: it runs until it hits the ceiling you chose, then waits for the next period, and the rest of the account carries on as normal.

It Hold Up, Even Against Clever Workarounds

A limit is only useful if it can't be stepped around, and a few details make sure of that.

Refused calls cost nothing. A call that's refused for a rate limit doesn't spend a task from the monthly allowance. The limit protects your total instead of eating into it.

Code mode is counted call by call. If a server uses code mode, an AI client can write one script that calls twenty tools. PopdockAI counts each of those twenty calls against your limits, and doesn't count the script on top. A script can't be used to get round a per-tool cap, and it isn't charged twice either.

The counts are exact. The counters are kept in the database, not estimated, so a limit of 60 per minute means 60, not "around 60 give or take."

Nobody gets locked out of signing in. Signing in and exchanging a token are never limited, so a user who hits their API limit can still get back into PopdockAI.

When to Use Rate Limits, and When Not to

Rate limits are a safety rail, not a billing tool, and they work best when you treat them that way.

Reach for a rate limit when:

  • You're opening an MCP server to a wider group and don't yet know how hard they'll use it.
  • One tool is expensive or slow on the source system, or writes data back, and should never be called in bulk.
  • An integration or script calls the Popdock API on a schedule and you want a ceiling in case it misbehaves.
  • One team's heavy use shouldn't be able to crowd out another's.

Leave it alone when:

  • You want to cap the account's total for the month. That's what the task allowance already does.
  • You're trying to stop someone seeing data. Rate limits control how often, not what. Use security on the list, the MCP server or the tool for that.
  • Your users make a handful of calls a day. A limit nobody will ever reach is just one more setting to maintain.

One practical note: changes take up to a minute to apply, because Popdock caches the rules close to where calls arrive. If you've just lowered a limit to stop a loop that's already running, give it a minute before you decide it didn't work.

Who Can Set Them

Limits on users and the account default are set by anyone who can manage security. Team limits need permission to update teams. Limits on MCP clients and MCP tools are part of PopdockAI's MCP features, so they're available on plans that include MCP. If you're not sure which plan your account is on, your Popdock administrator can check it on the Subscription tab.

The Short Version

Your monthly task allowance controls how much your account can do. Rate limits control how fast any one person, team, AI client or tool can do it, per minute, hour or day. Every rule that matches a call applies, refused calls cost nothing, and code mode scripts are counted call by call, so the limit holds. Set one wherever a single runaway agent or script could otherwise spend what everyone else was counting on.

There are five places a rate limit can live, and they all apply at once. Which one to use for which problem is the subject of the next article.

‍

‍

‍

From AI Potential to AI Performance in Minutes

One layer.
All your apps.
‍Any AI tool.

Join forward-thinking companies that are transforming their operations with intelligent, secure AI automation.

Try PopdockAI Free

See Documentation