Wokwi
Also known as: Wokwi CI, Wokwi for VS Code, Wokwi Simulator
An embedded and IoT simulator with a CI runner: Wokwi simulates ESP32, Arduino, STM32 and Pico boards, and its CLI ships an experimental MCP server for coding agents.
Wokwi is a browser based simulator for embedded and IoT projects, and Wokwi CI runs the same simulation engine headless in a pipeline. It simulates Arduino AVR boards, ESP32 chips on Xtensa and RISC-V, STM32 parts and the Raspberry Pi Pico, with displays, sensors, motors and simulated WiFi that can reach the internet. Custom parts can be written in C or other WebAssembly languages through the Chips API, which is in beta.
Wokwi for VS Code brings the simulator into the editor for Zephyr, PlatformIO, ESP-IDF, Pico SDK, NuttX, Rust, Arduino CLI and MicroPython projects, with a GDB server for stepping through the simulated firmware.
For coding agents, the Wokwi CLI ships an experimental MCP server that runs over stdio with a CI token. Its tools start, stop, resume and restart a simulation, read and write the serial console, read pins, set sensor controls, take screenshots and export logic analyzer traces. In CI, a run passes or fails on expected serial text, and YAML automation scenarios press virtual buttons, change sensor values and compare screenshots.
Plans run from a free Community tier for open source projects to Wokwi Pro at €20 per seat a month billed yearly, which adds 2,000 CI minutes a month, offline use in VS Code and team billing. On premises deployment of the CI server is available by contacting Wokwi. Wokwi B.V. runs the service.
Vendor details
Canonical URL
https://wokwi.com
Category
Agent infrastructure
Subcategory
Embedded and IoT simulator with CI and an MCP server
Funding status
Wokwi B.V. runs the service. No funding is stated on its site.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Coding agents connect through the MCP server in the Wokwi CLI, and Wokwi's docs name Copilot, Claude Code, Cursor, Gemini and ChatGPT as clients. CI runs through the CLI and the wokwi-ci-action for GitHub Actions, with PlatformIO, ESP-IDF and Pico SDK projects detected automatically. Wokwi for VS Code and JetBrains IDEs runs simulations inside the editor, and a Private IoT Gateway on a paid plan connects simulated devices to the local network.
In practice
Your coding agent rewrote the ESP32 firmware for a sensor node. Through the Wokwi MCP server it starts a simulation, sets the sensor's humidity, reads the serial console and takes a screenshot of the display.
Your team wants every commit tested without a board on the desk. Wokwi CI runs the firmware in the cloud simulator and fails the build when the expected serial text never appears.
You are prototyping a circuit before ordering parts. The browser simulator wires up the board, display and sensors, and custom parts can be modeled through the Chips API.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
3.5 / 14 capabilities · 25%
| Integrations & Tool Calling | Not documented |
|---|---|
|
Wokwi works inside its own simulator and does not reach into other software on an agent's behalf. Its MCP tools start and stop simulations, talk to the serial monitor, read pins, set part controls and capture screenshots, all within the running project. A simulated ESP32 can join the Wokwi-GUEST WiFi network and send MQTT, HTTP, HTTPS or WebSocket traffic out to the internet. Paying users can route that traffic through a Private IoT Gateway on their own computer, which also lets the firmware reach services on the local network and accept incoming connections. That traffic comes from the firmware under test. Wokwi offers no connectors for an agent to act in outside services. Sourcedocs.wokwi.com/wokwi-ci/mcp-supportread 2026-10-08 |
|
| Workflow Orchestration | Not documented |
|
Each Wokwi run is one simulation that the caller starts, stops, resumes or restarts. Automation scenarios lay out a fixed sequence for a single board: set a control, wait, check serial output, compare a screenshot. Teams with long suites split them into several CI runs. Wokwi does not coordinate several agents or chain work from one run into the next. That sequencing stays with the calling agent or the CI pipeline. Sourcedocs.wokwi.com/wokwi-ci/automation-scenariosread 2026-10-08 |
|
| Knowledge Grounding & RAG | Not documented |
|
Wokwi stores projects, wiring diagrams and parts. Its MCP server hands an agent the project's wokwi.toml and diagram.json files as readable resources, so the agent can see which firmware is loaded and how the board is wired, and a simulation start returns any diagram lint warnings. Those files describe the board under simulation. Wokwi keeps no knowledge base over a customer's documents and has no assistant that answers questions from one. Sourcedocs.wokwi.comread 2026-10-08 |
|
| Human Oversight & Guardrails | Not documented |
|
Agents call the MCP tools directly, and each call takes effect straight away, whether it starts a simulation, changes a sensor value or writes to the serial port. Wokwi has no step where a person reviews or approves what an agent is about to do. The MCP server itself is still labeled experimental in the docs and in the CLI's README. Sourcedocs.wokwi.com/wokwi-ci/mcp-supportread 2026-10-08 |
|
| Security, Identity & Governance | Not documented |
|
Access rests on Wokwi accounts and API tokens. A CI token is created in the Wokwi CI Dashboard, starts with wok_, runs to 44 characters and is passed to the CLI, the GitHub Action or the MCP server as an environment variable or repository secret. The VS Code extension is licensed by signing in to a Wokwi account from the editor. Wokwi Pro adds per seat billing, team invoices and unlisted projects. Wokwi does not offer single sign on, user roles or team permission settings, and it publishes no security page, trust center or certification. Sourcedocs.wokwi.com/wokwi-ci/cli-usageread 2026-10-08 |
|
| Observability & Auditability | Not documented |
|
Each run leaves evidence about the firmware: the serial log, which the CLI can save to a file, logic analyzer captures exported as VCD files, and screenshots of display parts taken at a set point in simulated time. CI time is metered against an allowance measured over a rolling 30 days, with a wall clock limit on each session. Wokwi keeps no audit trail of which user, token or agent started a run or what it changed. Sourcedocs.wokwi.com/wokwi-ci/getting-startedread 2026-10-08 |
|
| Memory & State Persistence | Partial |
|
Within one MCP session, stopping a simulation pauses it, and the resume tool picks it up where it left off; restart begins again from scratch. Serial output builds up in a buffer the agent can read with or without clearing it. A simulated microSD card can be preloaded from a folder or an image and written back when the run stops or restarts, so the next run starts with the files the firmware left behind. All of that is the board's own state. Wokwi keeps no record of an agent's earlier sessions for it to draw on. Sourcegithub.com/wokwi/wokwi-cliread 2026-10-08 |
|
| Deployment & Data Residency | Partial |
|
Simulations run on Wokwi's cloud by default. Wokwi CI uses a stateless server that runs simulations in parallel and deletes the firmware after each one, and on premises deployment is available by contacting Wokwi. On the Pro plan, the VS Code extension can also run offline from a simulator engine stored on the developer's machine, refreshing it whenever it next goes online. For machines with no internet access at all, Wokwi prepares a custom offline installation package on quote. The cloud service offers no choice of hosting region. Sourcedocs.wokwi.com/vscode/offline-moderead 2026-10-08 |
|
| Prebuilt Agents / Templates / Packs | Partial |
|
A lot comes ready to run. The simulator supports about 17 microcontroller types: AVR boards such as the Arduino Uno, Mega and Nano plus the ATtiny85, ten ESP32 chips across Xtensa and RISC-V, three STM32 parts and the Raspberry Pi Pico. Around 66 documented parts sit alongside them, from sensors, displays and motors to a logic analyzer and a microSD card. Wokwi also publishes starting projects, such as an ESP-IDF hello world repository, example custom chips (an EEPROM, OLED and LCD drivers, a logic analyzer with protocol decoding) and community projects on wokwi.com. These are boards and projects to build from; Wokwi offers no prebuilt agents. Sourcedocs.wokwi.com/getting-started/supported-hardwareread 2026-10-08 |
|
| Triggers & Channel Coverage | Not documented |
|
A simulation starts when someone opens a project in the browser or in VS Code, when the CLI runs, or when an agent calls the MCP server. In CI, the wokwi-ci-action step runs after the firmware build inside the customer's own GitHub workflow, on whatever events that workflow already listens to. Wokwi itself has no scheduler, webhook or event listener that starts a simulation on its own. Sourcedocs.wokwi.com/wokwi-ci/getting-startedread 2026-10-08 |
|
| Model Flexibility & Routing | Not documented |
|
No AI model runs inside Wokwi. The coding agent that connects to the MCP server brings its own model, and the docs name Copilot, Claude Code, Cursor, Gemini and ChatGPT as clients. Wokwi has no setting for choosing a model. Sourcedocs.wokwi.com/wokwi-ci/mcp-supportread 2026-10-08 |
|
| APIs / SDKs / MCP Extensibility | Full |
|
A command line tool and an SDK open the simulator to code. With a CI token, the Wokwi CLI runs projects described in wokwi.toml and diagram.json, lints diagrams for unknown parts and bad pin connections, exports logic analyzer traces, preloads SD cards and starts a GDB server. Running wokwi-cli mcp starts the MCP server over stdio. The open source CLI defines its eleven tools for starting, stopping, resuming and restarting a simulation, checking status, reading and writing serial data, reading pins, setting part controls, taking screenshots and exporting VCD files, plus the project's two config files as resources. The Chips API, in beta, is an SDK for writing custom parts in C or another language that compiles to WebAssembly, with calls for GPIO, analog, timers, UART, I2C, SPI, attributes and framebuffers, and experimental Verilog support. Sourcedocs.wokwi.com/chips-api/getting-startedread 2026-10-08 |
|
| Testing, Debugging & Optimization | Full |
|
A firmware build becomes a pass or fail check. The CLI fails a run when expected serial text never appears or when forbidden text does, and exits with a set code, 42 by default, when the simulation times out. YAML automation scenarios press buttons, change sensor values, check pin levels, wait for serial output and compare a display with a saved screenshot, though Wokwi still labels scenarios alpha. The wokwi-ci-action runs these checks in GitHub Actions after each build. On ESP32 boards, an instruction coverage file records which addresses ran and how often each branch was taken, and a GDB server lets a developer step through the simulated firmware from VS Code or the command line. Sourcedocs.wokwi.com/wokwi-ci/cli-usageread 2026-10-08 |
|
| Browser / Computer-use | Not documented |
|
Agents reach Wokwi through MCP tool calls and the CLI rather than through a screen. The browser simulator and the VS Code extension are built for people, and the screenshot tool captures a simulated display part such as an OLED, not a desktop. No browser or desktop session is operated by the agent. Sourcedocs.wokwi.com/wokwi-ci/mcp-supportread 2026-10-08 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Community free · Hobby €5.6/mo · Hobby+ €8.1/mo · Pro €20 per seat/mo, all billed yearly · Classroom by quote
Per account on Hobby and Hobby+, per seat on Pro, with CI simulation minutes capped per rolling 30 days
Included quota
CI minutes per rolling 30 days: Free 50, Hobby 200, Hobby+ 500, Pro 2,000. Each CI session has a wall clock limit of 5 minutes on Free and Hobby, 10 on Hobby+ and 50 on Pro. Fast build minutes: Hobby 100, Hobby+ 500, Pro 1,000.
What is public
Wokwi publishes four plans at yearly rates: Community free for open source, Hobby at €5.6 a month, Hobby+ at €8.1 a month and Pro at €20 per seat a month. CI minute allowances per plan are in the docs. Classroom licenses and on premises deployment are by request.
Billing mechanics
Plans are billed yearly at the rates shown, with Pro charged per seat. CI usage draws from a simulation time allowance measured over a rolling 30 days.
Cost watchouts
An agent that runs a simulation for every change spends CI minutes quickly, and Free and Hobby cap each CI session at 5 minutes. Only Pro lists a large CI allowance, and it is priced per seat. Monthly billing prices are not published.
Variable cost rationale
Plans are flat yearly prices, per seat on Pro, and CI time is a capped allowance rather than a meter, so the bill does not grow with how often an agent runs simulations. The limit shows up as waiting for minutes, not as a larger invoice.
Additional watchouts
Size the plan by CI minutes and session length rather than by price, since agent loops consume simulation time fast.
Overage / add-ons
No overage rate is published. Simulation time beyond the CI minute cap waits for the rolling 30 day window, which does not reset on a fixed date.
Sales call required
Mixed (some tiers require a call)
Free / trial
Community plan free for open source projects, with unlimited simulations and public projects and 50 CI minutes a month
Lowest paid plan
Hobby at €5.6 a month billed yearly
Commercial notes
Wokwi B.V. runs the simulator, which is used in universities and by makers as well as in commercial firmware teams. The CLI and its MCP server are open source under the MIT license.
Key ambiguities
Only yearly rates are shown, so the monthly price is not known, and the page does not say whether MCP use draws on the same CI minutes or how on premises deployment is priced.
Missing data
Monthly billing prices, on premises pricing, Classroom pricing and cancellation terms.
Related vendors
- AgentOps — Agent observability and debugging platform: open source SDKs trace…
- Agno — Python agent framework and AgentOS runtime (formerly Phidata) for…
- AIsa — Resource and payment gateway for AI agents, with one key to 110+…
- AlphaBitCore — AI control plane for regulated financial firms: one gateway enforces…
- Anchor Browser — Cloud hosted browser infrastructure that lets AI agents operate real…
- Apify — Cloud platform and marketplace of more than 73,000 ready-to-run…