What a Day of Figma Costs an Agent

Is two hundred Figma calls a day a lot? The better question is whether it's enough for a class.

Three terms first. An MCP call: one request an agent sends to Figma, to write, render, or read. The rate limit: on Education, 200 calls a day and 10 a minute. AI credits: Figma's separate budget for its own AI features, 3,000 a month.

Overnight, eight agents ran experiments on Bear's Figma for Education account and logged every call: 61 rows. This film reads that ledger. The common job across experiments was build, then read back — one call makes the file, one write builds it, a render shows it, a few reads pull its data. Across four experiments: six to ten calls each. A FigJam table, drawn and read back, took four. A screen, built and audited, took eleven with setup. The range ran from three, for a Weave access check, to thirteen, for a set of tutor states.

In all, 61 calls, and the tool used most was the render — seventeen times, under a third of two hundred. Figma's limits apply only to tools that read data — making files is exempt, and so are some writes — so counting only reads brings the total to 35.

Some arithmetic, not a measurement: an average experiment took about 7.5 calls. Twenty students sharing one login, one run each, would use about 150. On their own accounts, each student could run the biggest experiment about fifteen times a day. The other cap, ten a minute, held too — the two busiest experiments made eight calls in a minute, close but under; all agents together logged up to 24 in one minute, thirteen of them reads, with no logged rate-limit error. Self-logged times, so approximate — one night can't say why.

The second cost is Claude: 21 headless Claude Code builds reported $3.27 at list price, under Bear's own login, not charged to a card. Six agents stopped around 3am on Bear's weekly Claude usage limit. Unmeasured entirely: Figma's AI credits — no agent ran a Figma AI feature, since the rules forbade it, and nobody read the credit meter; that measurement is still pending Bear.

Is the budget enough for real classwork? On one night's ledger: strong for students on their own accounts — the biggest experiment used thirteen of two hundred. But the agents were capped at fifteen calls or fewer, one skipped checks to stay under it, and Figma says MCP writes will eventually become a usage-based paid feature.

Your Turn: Paste into Claude: "Here's my plan for a Figma experiment with an agent: [paste your plan]. Before I start, list every Figma MCP call it needs (files, writes, renders, reads) with a total, check it against 200 a day and 10 a minute, and flag any write I won't be able to see." Then check two things yourself: does a render follow every write you need to see, and afterward, is your real count close to the plan?

Sources: developers.figma.com/docs/figma-mcp-server (rate limits and access; tools and prompts); help.figma.com/hc/en-us/articles/32132100833559 (Guide to the Figma MCP server); 37611757713431; 42614902212887 (AI credit updates FAQ); 33459875669015 (How AI credits work). Every factual claim checked against these sources and the experiment record before rendering. An agent ran the experiment on Bear's Figma for Education account; Bear reviewed the film from an educational perspective. An independent series, not affiliated with or endorsed by Figma.

0:00 The question and terms 0:21 Eight agents, 61 calls 0:36 The common job: build, then read back 0:49 A table, a screen, and what they cost 1:01 The tool used most: the render 1:15 Arithmetic for a class of students 1:29 The other cap: ten a minute 1:47 The Claude side: 21 headless builds 2:06 Unmeasured: Figma's AI credits 2:22 Is the budget enough? 2:41 Bear's review 3:16 Your turn

Series: Figma for Educational AI. @NikBearBrown

SEO Keyword Tags

Figma MCP rate limits, Figma Education MCP calls, Figma AI credits, Figma MCP server budget, Claude Code Figma cost, Figma classroom AI usage, Figma MCP calls per day, Figma render tool limit, Figma education plan capacity, agentic AI cost tracking, Figma MCP tools and prompts, Claude Code headless builds cost, Figma rate limit per minute, Figma AI credit meter, classroom AI budget planning

What this video covers

• 0:00 — The question and terms 0:21 Eight agents, 61 calls 0:36 The common job: build, then read back 0:49 A table, a screen, and what they cost 1:01 The tool used most: the render 1:15 Arithmetic for a class of students 1:29 The other cap: ten a minute 1:47 The Claude side: 21 headless builds 2:06 Unmeasured: Figma's AI credits 2:22 Is the budget enough? 2:41 Bear's review 3:16 Your turn

Transcript

A lightly formatted transcript of the video, with timestamps so you can jump to any section on YouTube.

The question and terms 0:21 Eight agents, 61 calls 0:36 The common job: build, then read back 0:49 A table, a screen, and what they cost 1:01 The tool used most: the render 1:15 Arithmetic for a class of students 1:29 The other cap: ten a minute 1:47 The Claude side: 21 headless builds 2:06 Unmeasured: Figma's AI credits 2:22 Is the budget enough? 2:41 Bear's review 3:16 Your turn

[0:00] Kiaora, this is Liam in for Bear. Is 200 Figma calls a day a lot? Better question, is it enough for a class? Three terms: an MCP call, one request an agent sends to Figma to write, render, or read. The rate limit on education, 200 calls a day and 10 a minute. And AI credits, Figma's budget for its own AI features, 3,000 a month. Last night at Bear's request, eight agents ran experiments on his Figma for education account and logged every call, 61 rows. An agent compiled this film from that ledger. Bear has since reviewed it from an educational perspective and it covers the overnight agents only. The common job, build one small design, a card or a button, then read it back. One call makes the file, one write builds it, a render shows it, and a few reads pull its data.

Across four experiments, six to 10 calls. A FigJam table drawn and read back took four. A screen built and audited took 11 with its setup. And the range ran from three for a weave access check to 13 for a set of two or states. In all 61 calls and the tool used most was the render, 17 times. That's under a third of 200. And Figma says its limits apply to tools that read data. Making files is exempt and so are some rights. Count only reads and it's 35. Now some arithmetic, not a measurement. An average experiment took about seven and a half calls. 20 students sharing one login, one run each, would use about 150. On their own accounts, each could run the biggest one about 15 times a day. The other cap is 10 a minute. All 61 calls fell inside

About 6 minutes. The two busiest experiments made eight calls in a minute, close but under. All the agents together logged up to 24 in 1 minute, 13 of them reads, and none logged a rate limit error. Those times are self-logged, so they're approximate. One night can't say why. The second cost is Claude. 21 headless Claude code builds reported $3.27 at list price under Bear's Claude login, not money charged to a card. The agents running the experiments reported no figure, and the knights report says six of them stopped around 3:00 in the morning on Bear's weekly clawed usage limit. What's unmeasured is Figma's AI credits. No agent ran a Figma AI feature since the rules forbade it and nobody read the credit meter. To measure the 3,000, Bear reads it before and after a make action. Figma says its agent in design is

Free during its beta. That half is pending Bear. So, is the education MCP budget enough for real classwork? On one night's ledger, strong for students on their own accounts. The biggest experiment used 13 of 200, but the agents were capped at 15 calls or fewer. One of them skipped checks to stay under it, and Figma says MCP rights will eventually be a usage-based paid feature. Bear reviewed this film after the overnight run from an educational perspective, and he agrees with its point. His note isn't in a chat thread. With no Figma file here, it sits in the experiment record next to the evidence. On the films that have a Figma file, it went into the file itself. And the review cost calls of its own, 22 rows tagged review in the same ledger. Seven notes written, four calls refused, and 11 reads,

Two of them checking what was written. If someone other than Bear had to look, a small script would tell them on Slack, Discord, or email. Here's its test run for another film's file, a link, not a copy of the note. No one else had to look at this one, so nothing was sent. Return, paste this into Claude. Here's my plan for Figma experiment with an agent, then paste your plan. Before I start, list every Figma MCP call needs, files, rights, renders, and reads with a total. Check it against 200 a day and 10 a minute, and flag any right I won't be able to see. Then check two things yourself. Does a render follow every right you need to see? And afterwards, is your real count close to the plan? What a day of Figma costs an agent at Nick Bear Brown.