All guides
Business operations · Angelcam MCP Connector

How busy was the
restaurant
?
See how busy your dining room gets.

Count guests from live or recording snapshots and see when your dining room is busiest.
Pick your assistant Open MCP Connector
Claude or ChatGPT or Grok Bot
Live or recording snapshots
Illustrative dining room camera view
Dining room
Illustrative example report with sample data
Example report
Camera scenes and report examples are illustrations.
How it works

From a camera picture to an answer you can act on.

Camera → Angelcam snapshots → AI count → daily report. Use live or recording snapshots to build a history of guest counts and see your busiest sampled periods.
  1. Your camera

    Dining room

    Use a camera connected to your Angelcam account.
  2. Angelcam

    Live or recording snapshots

    Choose a current image or a snapshot from an earlier recording through the MCP Connector.
  3. Your assistant

    How many guests were there?

    Use the guide’s prompts to inspect what is visible.
  4. You

    Counts you can review

    Use the frames, count history and edition-specific reporting steps to review what the assistant found.
Choose your snapshot source

Check now. Or look back.

Every use case works with live snapshots and snapshots from recordings. Choose the source that fits the period you want to check.
Check from now on

Live snapshots

Count the room during service, save each observation and build the day’s report from the samples you collected.
No Cloud Recording required.
Use live snapshots with your assistant
Prompt
Use live snapshots for the guest-count guide. During service, take a fresh Angelcam snapshot of “Dining room” at each scheduled check. First confirm a sampling interval supported by my assistant. Count guests and staff separately, excluding the patio and reflections. Save the camera, timestamp, counts and uncertainty in the configured log without duplicate rows. At the end of service, report the busiest sampled periods and missing checks. Never backfill missed live checks with invented counts or present the sum of snapshots as unique daily visitors.
Copy
Look back at earlier times

Recording snapshots

Count snapshots from a past service period, then compare busy and quiet times without running checks during the day.
Requires Cloud Recording and footage for the selected period.
Use recording snapshots with your assistant
Prompt
Use recording snapshots for the guest-count guide. Review “Dining room” for the day I specify, from 11:00 through 23:00 Europe/Prague, one frame every 15 minutes, in batches supported by Angelcam. Count guests and staff separately using our room description. Save the camera, original timestamp, counts and uncertainty in the configured log without duplicate rows. Report busy periods and missing frames. Do not treat sampled occupancy counts as unique daily visitors.
Copy
Live checks build a log from the time they start. They cannot recover missed checks. Recording snapshots can review earlier footage within your retention period. Check the scheduling intervals and storage available in your assistant.
Before you start

Try a frame you can count yourself.

Compare the assistant’s count with the picture. Tell it which areas show staff, guests or people you want excluded.
Illustrative dining room view for checking the prompt
Pick your assistant

One use case. Choose your way to build it.

The steps and connected apps differ. Follow the edition for the assistant you use.
Anthropic

Claude

Use the Claude edition for its own connection steps, prompts and reporting workflow.
Before you start
  • An Angelcam account with a camera that clearly shows the scene. All use cases support live snapshots and recording snapshots. Cloud Recording is required only for snapshots from recordings; the live option does not need it.
  • A paid Claude plan (Pro, Max, Team or Enterprise).
  • Google Sheets connected to Claude, under Customize → Connectors. If your rota is in a sheet too, the dashboard can put staff next to guests.
Set it up in Claude
OpenAI

ChatGPT

Use the ChatGPT edition for its own connection steps, prompts and reporting workflow.
Before you start
  • An Angelcam account with a camera that clearly shows the scene. All use cases support live snapshots and recording snapshots. Cloud Recording is required only for snapshots from recordings; the live option does not need it.
  • A paid ChatGPT plan that can add custom MCP servers, plus ChatGPT on the web at chatgpt.com. For now, they can be added only on the web. On Business, Enterprise and Edu, an admin usually adds the server for the workspace, and you install it.
  • One free slot for a scheduled task. Plus has 5, Business 10, and Pro and Enterprise 15.
Set it up in ChatGPT
xAI

Grok Bot

Set up a Grok Bot routine, save a CSV log and build a report you can review.
Before you start
  • An Angelcam account with a camera that clearly shows the scene. All use cases support live snapshots and recording snapshots. Cloud Recording is required only for snapshots from recordings; the live option does not need it.
  • Grok Bot with access to custom MCP connections and routines. Choose one Bot to own this workflow.
  • Optional: Slack for the morning summary. A CSV log in the Bot’s workspace is enough to start; no spreadsheet connection is required.
Set it up in Grok Bot
Step by step

From connection to a result you can review.

Both snapshot sources work. The walkthrough below uses recording snapshots. Use the source prompts above for the other option.
ClaudeChatGPTGrok Bot

The idea

You know how much you sold. You probably don’t know when the room was full and when it sat half empty. Your dining room camera has the answer.
In this guide Angelcam keeps the day’s recording of your dining room and hands Claude a frame every 15 minutes. Claude counts the guests in each one, every night after closing, and builds a dashboard you can open each morning. You set it all up in Claude, in about 20 minutes.

Who does what

Who does what: your dining room camera, any of 200+ brands, connects to Angelcam. Angelcam keeps the day in Cloud Recording, hands out recorded frames up to 12 per request, and gives Claude secure access with the MCP Connector. Claude counts guests and staff in each frame, runs every night at 23:30 and builds the dashboard. The results land in Google Sheets and a Claude artifact.
Angelcam delivers the camera and the connection. Claude does the analysis.
1

Connect Angelcam to Claude

In Claude, open
Customize
→
Connectors
, choose
Add custom connector
, paste
https://mcp.angelcam.com/readonly
and click
Add
. Then click
Connect
, sign in with your Angelcam account and approve, with “Pictures from your cameras” ticked.
2

Count the room once

Prompt
Find my dining room camera. How many people are in the dining room right now? Count guests and staff separately, and leave out the patio.
Copy
A Claude chat: Claude takes a live snapshot of the Dining room camera and counts 22 guests and 3 staff, leaving out the people on the patio.
Tell Claude what not to count: the patio, staff, a reflection in the window. Illustration.
Check the number against the picture yourself. If Claude keeps counting the wrong thing, add one sentence about your room, for example “The bar staff stand behind the counter on the left.”
3

Count a whole day from the recording

Prompt
Go through yesterday’s recording of the dining room from 11:00 to 23:00, one frame every 15 minutes, and count the guests in each frame.
Copy
That is 49 frames. Angelcam hands them out in batches of up to 12, so Claude asks five times, and the whole day takes a couple of minutes.
A Claude chat: Claude reads the Cloud Recording in batches of 12 frames and lists the guests and staff for each 15-minute frame.
One row per frame: time, guests, staff and a note. Illustration.
4

Do it every night

Create a Google Sheet called
Guest counts
with five columns: Date, Time, Guests, Staff and Note. Then ask Claude for a scheduled task:
Prompt
Every day at 23:30, count today’s dining room guests every 15 minutes from 11:00 to 23:00, using the Angelcam recording of “Dining room”. Count guests and staff inside the dining room separately, and leave out the patio. Times are Prague time. Add one row per frame to the Guest counts sheet.
Copy
Claude proposes a scheduled task named Daily guest count: daily at 23:30, using Angelcam and Google Sheets, with a Schedule button.
Click Schedule. The task runs in the cloud every night. Illustration.
5

Build the manager dashboard

Prompt
Build an artifact called Dining Room Pulse. Each time I open it, read the Guest counts and Staff sheets. Show cards for today’s peak, busiest hour, guests per staff member at the peak and the afternoon low; a bar chart of guests every 15 minutes with the average of the last four same weekdays and the staff capacity at 8 guests per person; a staffing table by hour with suggestions; a frame viewer that fetches the recorded frame from Angelcam; a weekday-by-hour heat map for the last four weeks; and what stands out.
Copy
The Dining Room Pulse dashboard: cards for a peak of 29 guests at 20:00, busiest hour 19:30, 9.7 guests per staff member at the peak and an afternoon low of 6; a bar chart of guests every 15 minutes against staff capacity; a staffing table with suggestions; a frame viewer; a four-week heat map; and Claude’s notes.
The manager dashboard, with sample data. The staffing table turns guest counts into rota decisions.
Open the live example of Dining Room Pulse ↗ (sample data).
Click a bar and the dashboard asks Angelcam for that recorded frame, so you can check any count with your own eyes.

Good to know

  • A count is one moment.
    It tells you how full the room was, not how many guests came that day. Your till is still the source for covers.
  • Any camera you have.
    Angelcam works with 200+ camera brands. Cloud Recording keeps the footage the count runs on.
  • Expect a small error.
    In a full room, people behind others or behind plants can be missed, so a count can be two or three off.
  • Recording retention
    decides how far back you can recount. The nightly task only needs the same day.
  • Privacy.
    Claude counts people. It does not recognise faces or identify anyone. Check your local rules on using cameras for analytics, and tell guests where required. The frames Claude looks at are sent to Anthropic as images. You can turn pictures off when you connect, but then Claude gets links only and can’t judge the scene itself.
  • Add your other tools.
    If your bookings live in a calendar or sheet connected to Claude, ask it to put them next to the counts: booked tables against guests actually seated.
  • It’s a beta.
    You may hit a rough edge. If something doesn’t work or you miss something, write to us in the chat in MyAngelcam.
More ideas: angelcam.com/builders. Setup for other assistants: mcp.angelcam.com.

The idea

You know how much you sold. You probably don’t know when the room was full and when it sat half empty. Your dining room camera has the answer.
In this guide Angelcam keeps the day’s recording of your dining room and hands ChatGPT a frame every 15 minutes. Every night after closing, ChatGPT counts the guests in each frame and sends you a short summary. At the end of the month, it turns the counts into a report. You set it all up in ChatGPT on the web, in about 20 minutes. Use Claude instead? There is a Claude version of this guide.

Who does what

Who does what: your dining room camera, any of 200+ brands, connects to Angelcam. Angelcam keeps the day in Cloud Recording, hands out recorded frames up to 12 per request, and gives ChatGPT secure access with the MCP Connector. ChatGPT counts guests and staff in each frame, runs every night at 23:30, and sums up the day and the month. Results arrive as a nightly summary by email and push, the counts in the task chat, and a 4-week report.
Angelcam delivers the camera, the recording and the connection. ChatGPT does the counting.
1

Connect Angelcam to ChatGPT

On chatgpt.com, open
Plugins
, click + and choose
Add custom MCP server
. Name it Angelcam, paste
https://mcp.angelcam.com/readonly
as the server URL and choose OAuth. Tick “I understand and want to continue”, click
Create as a plugin
and install it. Then sign in with your Angelcam account and approve, with “Pictures from your cameras” ticked. ChatGPT calls this a plugin, but it’s a private connection in your own account. Angelcam doesn’t publish an official ChatGPT plugin.
Counting only reads, so this guide uses Angelcam’s read-only address. A connection made with it can never start or stop a recording.
An Add custom MCP server dialog in ChatGPT with the name Angelcam, the server URL https://mcp.angelcam.com/readonly, OAuth authentication and an “I understand and want to continue” checkbox.
Adding the Angelcam server in ChatGPT. Illustration: the labels in ChatGPT may differ slightly.
2

Count the room once

Start a new chat. Type @Angelcam at the start of your message and ask:
Prompt
Use Angelcam to find my dining room camera. How many people are in the dining room right now? Count guests and staff separately, and leave out the patio.
Copy
A ChatGPT chat: ChatGPT calls Angelcam for a live snapshot of the Dining room camera and counts 22 guests and 3 staff, leaving out the people on the patio.
Tell ChatGPT what not to count: the patio, staff, a reflection in the window. Illustration.
Check the number against the picture yourself. If ChatGPT keeps counting the wrong thing, add one sentence about your room, for example “The bar staff stand behind the counter on the left.”
3

Count a whole day from the recording

Prompt
Use Angelcam to go through yesterday’s recording of the dining room from 11:00 to 23:00, one frame every 15 minutes, and count the guests in each frame.
Copy
That is 49 frames. Angelcam hands them out in batches of up to 12, so ChatGPT asks five times.
A ChatGPT chat: ChatGPT reads the Angelcam Cloud Recording in batches of 12 frames and lists the guests and staff for each 15-minute frame.
One row per frame: time, guests, staff and a note. Illustration.
4

Do it every night

In the same chat, describe the job. ChatGPT turns it into a scheduled task:
Prompt
Every day at 23:30, use Angelcam to count today’s dining room guests every 15 minutes from 11:00 to 23:00, from the recording of “Dining room”. Count guests and staff inside the dining room separately, and leave out the patio. Times are Prague time. Write the counts as a table: time, guests, staff and a note. Then notify me with the peak, the quietest stretch, and any frame with more than 8 guests per staff member.
Copy
Check the schedule and the instructions, then open
Scheduled
in the sidebar to see the task and its next run. It runs on OpenAI’s side, so it keeps going when your computer is off. Exact times like 23:30 need a paid plan.
ChatGPT confirms a scheduled task named Daily guest count: daily at 23:30, four instructions, uses Angelcam.
The scheduled task. Illustration.
Name Angelcam and your time zone in the prompt. In ChatGPT, choosing an app applies to one message only, and Angelcam reports times in UTC.
5

Read the summary each morning

Under
Settings
→
Notifications
, choose Push, Email or both for tasks. Every night you get the peak, the quiet stretch, and the frames where the room had more guests than your team could serve.
An email from ChatGPT at 23:41, Daily guest count: peak 29 guests at 20:00 with 3 on the floor, 9.7 guests per person, above the target of 8; quietest 6 guests from 15:45 to 16:00; the frames over target; 49 frames counted from the Angelcam recording.
The nightly summary, with sample data. Illustration.
6

Get the 4-week report

Each night’s counts stay in the task’s chat. After four weeks, open that chat and ask:
Prompt
Make a report from this task’s results: yesterday in detail and the last 4 weeks. Show cards for yesterday’s peak, busiest hour, guests per staff member at the peak and the afternoon low; a bar chart of guests every 15 minutes with staff capacity at 8 guests per person; a staffing table by hour with suggestions; how each weekday compares; and what stands out. Give me all the counts as a spreadsheet.
Copy
ChatGPT’s report for the dining room: cards for a peak of 29 guests at 20:00, busiest hour 19:30, 9.7 guests per staff member at the peak and an afternoon low of 6; a bar chart of guests every 15 minutes against staff capacity; a staffing table by hour; a note that Fridays and Saturdays peak about 20% higher; and a spreadsheet to download.
The report, with sample data. The staffing table turns guest counts into rota decisions. Illustration.
The report shows the counts as they stood when you asked. For the next one, ask again in the same chat. To check a count with your own eyes, ask ChatGPT to show you the recorded frame from Angelcam.

Good to know

  • A count is one moment.
    It tells you how full the room was, not how many guests came that day. Your till is still the source for covers.
  • Any camera you have.
    Angelcam works with 200+ camera brands. Cloud Recording keeps the footage the count runs on.
  • Expect a small error.
    In a full room, people behind others or behind plants can be missed, so a count can be two or three off.
  • Recording retention
    decides how far back you can recount. The nightly task only needs the same day.
  • Keep the task’s chat.
    If you delete or archive it, the task pauses. A task can also pause after a long quiet stretch, so look at
    Scheduled
    now and then.
  • Privacy.
    ChatGPT counts people. It does not recognise faces or identify anyone. Check your local rules on using cameras for analytics, and tell guests where required. The frames ChatGPT looks at are sent to OpenAI as images. You can turn pictures off when you connect, but then ChatGPT gets links only and can’t count anything.
  • Add your other tools.
    If your bookings live in an app connected to ChatGPT, ask it to put them next to the counts: booked tables against guests actually seated.
  • It’s a beta.
    You may hit a rough edge. If something doesn’t work or you miss something, write to us in the chat in MyAngelcam. If you connected the full address, https://mcp.angelcam.com, you can also ask ChatGPT to send feedback to Angelcam.
More ideas: angelcam.com/builders. Setup for other assistants: mcp.angelcam.com.

The idea

Your dining room camera already records the day. Angelcam provides the frames; Grok Bot counts the visible guests, saves a record and prepares a morning report. Start with a count you can check, then repeat the same method each night.

Who does what

Angelcam keeps the recording and delivers the frames you request. Grok Bot separates guests from staff, records the counts in guest-counts.csv and builds a report from those rows. You review the pictures and decide what the numbers mean for your restaurant.
1

Connect Angelcam to Grok Bot

In the Bot that will own the check, ask:
Prompt
Add a custom MCP server named Angelcam at https://mcp.angelcam.com/readonly using streamable HTTP. Start its OAuth sign-in, then list the tools it offers. Do not change any cameras or recording settings.
Copy
Complete the Angelcam sign-in in your browser and allow camera pictures. Keep passwords out of chat. Confirm the Bot can access the correct camera before continuing. The read-only connection supplies camera information and images; Grok Bot interprets them.
2

Count the room once

Prompt
Find my Angelcam camera “Dining room”. Take a current snapshot and count guests and staff separately. Exclude the patio and reflections. Say which areas are unclear and show me the image so I can check the count.
Copy
Illustrative dining room camera view with faces blurred
Check the count against the picture. Illustrative scene; faces blurred.
Correct the areas the Bot should include or exclude. Save that description with the workflow; a busy bar, a mirror or an outdoor table can otherwise change what it counts.
3

Review one complete day

Prompt
Use the Angelcam recording of “Dining room” for yesterday, from 11:00 through 23:00 Europe/Prague, one frame every 15 minutes. Request frames in the batches the connector permits. Count guests and staff separately using our room description. Save guest-counts.csv with Date, Local time, UTC time, Guests, Staff, Note and Camera. Use “unclear” instead of a made-up count when the image is unavailable or obstructed. Do not duplicate a camera and timestamp already in the file.
Copy
Compare several rows with their recorded frames. Keep missing frames visible in the report rather than treating them as an empty restaurant.
4

Create the nightly routine

Prompt
Create a routine called Dining room guest count, every day at 23:30 Europe/Prague. Review today’s recording of “Dining room” from 11:00 through 23:00 at 15-minute intervals, using the counting method we tested. Update guest-counts.csv without duplicate rows. Report the peak count, busiest sampled hour, missing frames and anything uncertain in this chat. If recording access fails, report the failure and never reuse another day’s pictures. Do not send messages outside this chat yet.
Copy
Adjust the hours to your actual service. Run after the final sample time and within the camera’s recording retention.
5

Test the complete routine

Open the Bot’s details and its Routines list. Check the saved schedule and time zone, then use Test with a known period or a safe camera state. A test uses real tools and can write to the log or post an alert.
Start with alerts going only to you. Check the timestamps, the saved rows, an unavailable camera, and a second run of the same period. Only enable team alerts after the results and permissions match what you intended. If the routine cannot read Angelcam unattended, pause it and reconnect before relying on it.
6

Build the morning report

Prompt
Build a page called Dining Room Pulse from guest-counts.csv. Show the peak and quietest sampled periods, counts over the day, missing or unclear samples, and comparisons with earlier matching weekdays when enough data exists. Include timestamps and the camera for each sample so I can request the recorded frame to check it. Refresh the page after each successful nightly count. Do not present a sum of snapshots as total daily visitors.
Copy
Illustrative restaurant report with blurred faces in the camera view
Example report layout with sample data. The Grok Bot page is created from your own log.
For a team summary, connect Slack and test one post to your chosen channel. Then explicitly add that channel and the summary you want to the routine.

Good to know

  • Counts are samples.
    A person can appear in several frames. This measures occupancy at a moment, not total covers or unique visitors.
  • People can be hidden.
    Check crowded frames yourself before making staffing decisions. The workflow counts visible people without identifying them.
  • Keep the log.
    Ask the Bot to save a CSV file in its workspace. Check that later runs can read it; recent run history alone is not your long-term record.
  • Picture access.
    Camera images are supplied to the AI service used by Grok Bot. Use only cameras and recipients you are authorized to include.
  • Review permissions.
    Connect Slack only if you want messages there. Check the destination and scope of any saved permission before enabling unattended posts. If permission is missing, the routine should report that and keep the result in chat.

Setup references

See Grok Bot skills and routines and routine settings and testing for current product controls.
Keep exploring

What else could your cameras
tell you
?

Use another guide as a starting point for your site.
See all guides Open MCP Connector
Camera scenes, screens and reports are illustrations or sample data.