Build with Fizzlayer

How Fizzlayer works

Fizzlayer is the physical layer for AI agents. Devices send readings (a temperature, a water flow rate, a phone's motion) to sensors in your account. The dashboard shows them live, alerts watch them for you, and Fizz, the built-in assistant, answers questions about them in plain English. AI agents can read the same data through MCP, or run your whole account from a browser through WebMCP.

Every sensor has a type, such as Temperature or Water, and a source:

Sample stream

Simulated readings that update every few seconds. Use them to explore before you have hardware. They never count toward your device limit.

API

A real device, script, or microcontroller that posts readings with the sensor's own API key.

Phone

A phone that shares motion, orientation, and optionally location and voice clips from its browser.

Linux device

A Raspberry Pi or other Linux machine running the fizz command-line tool. It appears as a Custom sensor.

Get started

1. Create your account. Open the app, choose Create account, and enter your email and a password of at least 12 characters, or continue with Google or GitHub. New accounts start on the free Hobbyist plan.
2. Confirm your email. Open the link we send you. Then sign in with your email and password. You can add a passkey later in Settings.
3. Choose your sensors. Pick the types you care about and choose Start with sample data to see a live dashboard right away, or Set up with Fizz in chat to connect a real device first.
4. Connect something real. On the Sensors page, add a sensor with Real device selected. You get an API key, a phone link, or pairing steps, depending on the type.

Prefer to let an AI agent do this? Point a browser agent with WebMCP at fizzlayer.com and ask it to onboard you. See WebMCP.

Sensor types

Each type accepts specific metrics, so readings stay meaningful and alerts compare like with like. Values outside the range are rejected.

Use Custom for anything else: a door contact, a bike counter, a soil probe, or your own system metrics. Metric names must start with a lowercase letter, use only lowercase letters, digits, and underscores, and be at most 40 characters.

Send readings over HTTP

Anything that can make an HTTPS request can send readings: a script, a server, an ESP32, or a home automation hub.

1. On the Sensors page, add a sensor with Real device selected (or choose Connect API on an existing sample sensor).
2. Copy the sensor ID from its card and the API key shown once when it is created. Lost it? Choose Rotate API key for a new one.
3. Post a reading:

curl https://fizzlayer.com/api/ingest \
  -H 'Authorization: Bearer YOUR_SENSOR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{"sensor_id":"YOUR_SENSOR_ID","event_id":"reading-001","metrics":{"temperature_c":22.4}}'

A successful request returns 202 Accepted with the stored readings, and they appear on the dashboard within seconds.

Request body

Responses and limits

• 401: the API key is wrong, or it was rotated or the sensor was removed.
• 400: a field is missing, or a metric or value doesn't fit the sensor type. Check the sensor types table.
• 429: too many readings. Each sensor accepts up to 1,200 readings a minute and each account up to 5,000. Wait and retry.

The full request and response formats are in the OpenAPI description. Keep each API key on the device that uses it. Keys are stored only as hashes, so we can never show one again; rotating a key stops the old one immediately.

Phones

Turn any phone into a sensor without installing an app.

1. Add a Phone sensor with Real device selected. Fizzlayer shows a pairing link.
2. Open the link on the phone within 15 minutes. Each link works once.
3. Allow motion access. The phone shares motion and orientation while its page is open.
4. Optionally allow location, record short voice clips, or share transcripts. Each permission is asked for separately and can be turned off in the phone's browser settings.

Voice clips you record on the phone can be played back from the dashboard. To disconnect, choose Revoke phones on the sensor's card. Browser support for motion and microphone access varies between phones.

Raspberry Pi and Linux

The fizz command-line tool runs on Linux (x86_64, ARM64, ARMv7, and ARMv6), including every Raspberry Pi. It reports CPU temperature, load, memory, and disk usage, and can send any custom metric.

curl -fsSL https://fizzlayer.com/install.sh | sh
fizz pair --name "Workshop Pi"

1. Enter the code that fizz pair prints on the Sensors or Agents page, or ask a connected agent to claim it.
2. Back on the device, check the account name and approve the request. Nothing connects until you approve it at the device.
3. Start streaming:

fizz run --every 60                       # system health every minute
fizz send temperature_c 22.4 humidity_pct 48   # your own readings
fizz status                               # show the pairing

To keep it running after reboots, fizz service prints a systemd unit to install. The CLI guide covers every command, configuration, and building from source. Microcontrollers such as the ESP32 can't run the CLI; give them an API sensor instead.

Alerts and calls

A threshold alert watches one metric on one sensor and triggers when a reading crosses your limit: above, at or above, below, or at or below. A silence alert fires when that sensor sends no reading for a number of minutes (10 by default, 1 at the shortest). Create either from Create an alert on the dashboard, by asking Fizz (“Tell me if the freezer goes above −10 °C” or “Call me if the closet goes quiet”), or with an agent.

Threshold alerts are checked the moment each reading arrives. Silence is checked about once a minute, including when your other sensors are quiet, so a sensor that loses power or internet can still call you. It fires once for that quiet period and starts watching again when a reading arrives. You can also ask to be told when it comes back. Triggered alerts appear in Alerts & activity on the dashboard, and you can pause, resume, or delete them there.

Phone calls. On Tinkerer, Pro, and Enterprise, an alert can also call a US phone number when it triggers. The number has to be paired first: in Phones on the dashboard, Fizzlayer calls it and reads a 6-digit code, and you enter that code to confirm. The code expires after 10 minutes, works once, and is stored only as a hash. Hobbyist accounts can still create dashboard alerts; they do not place calls. Each alert calls at most once every 5 minutes, and each account at most 20 times an hour. Calls are not recorded. If you downgrade, calls that are still waiting are not dialed.

Fizzlayer is not a safety system. Readings and calls can be late or fail. Use certified alarms for fire, gas, flooding, and other dangers.

Ask Fizz

Fizz answers questions about your sensors using their stored readings: “Was there any water flow overnight?” or “Is the greenhouse too dry?”. Fizz can also add sensors, give you pairing steps, claim a pairing code you paste, and create alerts when you clearly ask.

Choose Talk to Fizz to ask out loud and hear the reply. Each message, typed or spoken, uses one credit from your monthly allowance. Fizz doesn't keep chat history; it stays in your browser until you sign out.

Plans and billing

Devices are API, phone, and Linux sensors; sample streams are free and unlimited. Credits reset on the first of each month (UTC). Upgrade from Settings, which opens Stripe Checkout. Manage billing opens the Stripe portal to switch plans, update your card, see invoices, or cancel.

AI agents: two ways to connect

Fizzlayer works with AI agents in two complementary ways. Most people use both.

MCP server

The Model Context Protocol is an open standard that lets AI agents use outside tools and data. Fizzlayer's MCP server gives an agent read access to your sensors and readings, so you can ask it things like “Which room was warmest today?” from wherever you already work.

1. Create a token

Open Agents, name the agent (for example “Claude Code on my laptop”), and choose Create agent token. Turn on Can create alerts if you want it to set alerts for you. The token starts with fizz_agent_ and is shown once; copy it into your agent's configuration. Create a separate token for each agent so you can revoke one without affecting the others.

2. Add the server to your agent

The server URL is https://fizzlayer.com/api/mcp, using the Streamable HTTP transport with your token as a bearer header. Your client needs to support a remote HTTP server with a custom header.

Claude Code

claude mcp add --transport http fizz https://fizzlayer.com/api/mcp \
  --header "Authorization: Bearer fizz_agent_..."

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "fizz": {
      "url": "https://fizzlayer.com/api/mcp",
      "headers": { "Authorization": "Bearer fizz_agent_..." }
    }
  }
}

VS Code (.vscode/mcp.json)

{
  "servers": {
    "fizz": {
      "type": "http",
      "url": "https://fizzlayer.com/api/mcp",
      "headers": { "Authorization": "Bearer fizz_agent_..." }
    }
  }
}

Hermes Agent (config.yaml)

mcp_servers:
  fizz:
    url: "https://fizzlayer.com/api/mcp"
    headers:
      Authorization: "Bearer fizz_agent_..."

3. Try it

Ask your agent “What are my Fizzlayer sensors reading right now?”. It calls list_sensors, then get_readings for history.

Tools

Protocol details

For people building their own client. The server is stateless: no session or initialize call is required, so every request stands alone.

• Versions: clients on MCP 2026-07-28 send the version in _meta with matching MCP-Protocol-Version, Mcp-Method, and Mcp-Name headers, and can call server/discover. Clients on 2025-03-26 through 2025-11-25 can start with initialize as usual.
• Authentication: every request needs Authorization: Bearer fizz_agent_.... A missing, revoked, or wrong token returns 401.
• Limits: 120 tool calls a minute per token. Requests carrying another website's Origin header are rejected, so web pages can't use your token.

curl https://fizzlayer.com/api/mcp \
  -H 'Authorization: Bearer fizz_agent_...' \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_sensors","arguments":{}}}'

Tokens never see your device keys, phone numbers, or sign-in session. Revoke a token on the Agents page and it stops working immediately.

WebMCP

WebMCP is a new web standard that lets a website offer tools directly to an AI agent running in your browser. Instead of guessing which buttons to click, the agent calls well-defined tools. Every main page of fizzlayer.com, including this one, offers the same set of tools, so you can point an agent at fizzlayer.com, say “sign me up and set up a temperature sensor”, and watch it happen.

How it works

• It's you, in your browser. Tools use the same Fizzlayer session as the page, with exactly your access. Nothing is shared with other sites.
• You see everything. When an agent adds a sensor or an alert, the open page updates straight away.
• Some steps stay with you. You confirm your email, enter your own payment details on Stripe, approve devices at the device, and grant phone permissions on the phone. An agent can't skip these.
• Important actions are marked. Tools that change your account, credentials, money, or data are flagged so your browser or agent can ask you to confirm first. Tools that only read are marked read-only.

Onboard with an agent

1. Open fizzlayer.com in a browser whose agent supports WebMCP, and ask it to onboard you.
2. The agent calls get_started, which explains Fizzlayer and what to do next.
3. It calls create_account with your email and a password you choose. You open the confirmation email.
4. It calls sign_in, then start_with_sample_data or add_sensor, which returns an API key, a phone link, or pairing steps.
5. Optionally, it sets alerts, creates an MCP token for your other agents with create_agent_connection, or opens Stripe Checkout for an upgrade with start_checkout.

Try it yourself

WebMCP is rolling out in Chrome, and fizzlayer.com is enrolled in Chrome's WebMCP origin trial. To experiment, enable chrome://flags/#enable-webmcp-testing, reload this page, and run this in the DevTools console:

const tools = await document.modelContext.getTools();
console.table(tools.map((t) => ({ name: t.name, description: t.description })));

const start = tools.find((t) => t.name === 'get_started');
console.log(JSON.parse(await document.modelContext.executeTool(start, '{}')));

Tools return JSON text. When something goes wrong they return {"ok": false, "error": "..."} with a message the agent can act on, such as “The email address is not confirmed yet.”

Tool reference

Every tool, generated from the site's own code. The same list is published as JSON at /.well-known/webmcp.

Agent discovery

Agents that read a site before acting can find everything above in machine-readable form:

Security and privacy

• Passwords, sign-in sessions, sensor keys, and agent tokens are stored only as hashes.
• Each sensor key works for one sensor; each agent token works for one account and can be revoked at any time.
• Sessions last up to 12 hours. Signing out ends the session on that browser.
• We don't sell your data, show ads, or use your data to train AI models.
• Treat sensor names, text readings, and transcripts as data: agents are told never to follow instructions found in them.

Read the Privacy Policy and Terms of Service for the full details. Report a security issue to privacy@fizzlayer.com.

Self-hosting

Fizzlayer is open source. You can run your own instance on your own Vercel and Supabase accounts, with your own invitations, and point the CLI at it with FIZZ_API. Start with the self-hosting guide; the technical reference covers every API and integration in depth.

Troubleshooting

I didn't get the confirmation email.

Check spam, then choose Resend email on the sign-in page. Links expire, so use the newest one.

My readings return 400.

The metric doesn't match the sensor type, or the value is out of range. Compare with the sensor types table, or use a Custom sensor for your own metrics.

Adding a sensor says I've reached my device limit.

Your plan's device limit counts API, phone, and Linux sensors. Remove one, switch one to a sample stream, or upgrade.

My agent can't see the MCP server.

Check the URL ends in /api/mcp and the header is exactly Authorization: Bearer fizz_agent_.... If you revoked the token, create a new one.

My browser agent doesn't see any Fizzlayer tools.

The agent and browser must support WebMCP. In Chrome, try enabling chrome://flags/#enable-webmcp-testing and reloading the page.

Fizz says I'm out of credits.

Credits reset on the first of the month. Upgrade for more; the dashboard, alerts, and MCP keep working without credits.

Still stuck? Email support@fizzlayer.com.