AI agents have made more than 19 million tool calls to Miro’s MCP server since February 2026, and the users behind them span product, design, marketing, operations, and project management.
When we launched Miro’s MCP server in February, we had AI coding tools in mind. Model Context Protocol (MCP), the open standard that AI tools use to connect with other apps, gave AI tools like Claude Code and Cursor a way to read the context teams keep in Miro and to generate new content right onto the board.
What started as a boon for engineering teams quickly caught on well beyond that function. Since February, AI tools have made more than 19 million tool calls to the server. (A tool call is a single request from an AI agent to Miro, such as reading a board or adding a diagram to one.) Every month since launch has topped the month before it, and 3 in 4 people now using Miro’s MCP server are in roles outside of engineering.
“The explosive user adoption of Miro’s MCP server is showing the true demand for bringing your agents closer to the work teams are actively doing. By leveraging MCP tools with Miro, teams can visualize complex ideas across multiple formats on a collaborative canvas. What’s most exciting is the growth we’re seeing across all disciplines, showing Miro is well positioned to be the collaboration layer in the agentic era.”
Jeff Chow, Chief Product & Technology Officer at Miro
By the numbers
Adoption took off quickly, and it’s still climbing:
- 34.9x growth in monthly tool calls from February to August
- 16.8x growth in monthly users over the same stretch
- 4.3x+ growth in usage since May
- ~6.3M tool calls in September, the biggest month yet
And the people using MCP are leaning on it harder each month:
- 85,000+ organizations used the server in August
- 108% more tool calls per active user in August than in February
Who’s using it: product, design, marketing, ops, and engineering teams
Engineers and developers are still a big part of the story, but now they have plenty of company: people in product, design, marketing, operations, and project management roles together outnumber engineering users roughly two to one.
The variety of AI tools they connect with has broadened, as well. Fifteen different AI clients now work with Miro through MCP. Claude Desktop and Claude Code led August with almost two-thirds (61%) of usage, and ChatGPT has become the second-largest client, with the number of users in September nearly tripling over August numbers (172% growth). Microsoft Copilot, Gemini Enterprise, and Grok Build are in the mix too.
Fincore Ltd. is one product team putting this to work:
“The Miro MCP server changed how our product team thinks and ships. It gives our PMs the freedom to be bold on the canvas knowing Claude will help turn that thinking directly into the requirements our developers build from. Better ideas reach engineering faster.”
Dominic Le Garsmeur, Chief Product Officer at Fincore
The canvas as input and output
The Miro canvas does double duty for AI tools: it’s the context they read from and the place they build. In August, 57% of MCP tool calls had an AI tool reading a board and acting on what it found (Miro-to-code). The rest used MCP to accelerate delivery, with agents building boards, diagrams, and layouts right on the canvas (code-to-Miro). And because teams work together on that same canvas, shaping what agents read and editing what they build, the team’s judgment is always in the mix.
The code-to-Miro movement has been a big focus of our recent product work:
“Nobody ever solved a hard problem with a wall of AI text,” he says. “To really think it through, you draw it out and talk about it.”
Charles Bédard-Porter, Lead Product Manager at Miro
Agents can now build slides, clickable prototypes, project timelines that switch to a Kanban view, clean diagrams, and workshop boards that include dot voting. “AI making a deck isn’t news,” Charles says. “But with Miro your deck lands somewhere your team can open it up, read it, discuss, and fix it together before anyone has to present it.”
And the first draft is just a starting point. Teams can ask their agent to change one part of the board and leave the rest as it is, which Charles calls “the difference between a generator and a collaborator.”
Why it matters: bring your own AI, and work on it together
Most AI work still happens one person and one chat window at a time. A product manager asks Claude to synthesize a dozen customer interviews, and the synthesis stays in that chat until they paste it into a doc, a Slack thread, or a slide, where it sits as a static copy the rest of the team can only read.
Miro’s MCP server lets agents build directly on the board, in the formats the team works in. So that same product manager can ask Claude to turn those interviews into a user journey map. A designer can have ChatGPT set up a workshop with voting built in. An engineer can have Cursor map a service as a diagram before a design review. Each result is visual, editable, and open to everyone working on the board. And the next agent that reads the board picks up the team’s edits and decisions along with the original output.
Those three people started with three different AI tools, and their work still ended up side-by-side on the same board, where teams work and decide together. That’s the idea behind the server: bring your own AI, or use ours, and pick the right model for each job without locking your work to a single provider. No matter which tool built it, the work lives in one shared workspace, with the same context for everyone.
What’s next
Everything we’re seeing with MCP adoption bears out the belief we’ve been building on all along: that AI’s output is more useful when the whole team can see it, shape it, and decide on it together. That idea is at the heart of the Intelligent Canvas™, and on October 20, we’ll show you where we’re taking it next.
Note: Figures are drawn from Miro’s internal product analytics and count tool calls made to Miro’s MCP server by connected AI clients between the server’s public launch to the time of publication, excluding usage by Miro employees.
