Model Context Protocol

Run your cameras and video from any AI agent

Connect Claude, Cursor, Codex, Gemini CLI, VS Code or any other MCP agent to Ceptory. Say what to watch for, and the agent sets up the camera workflow, checks alerts, follows movement across cameras and searches your footage for you.

How it works

Three steps, about two minutes

  1. 1

    Click Connect

    One click adds Ceptory to Claude, Cursor or VS Code. For Codex, Claude Code and others, paste https://mcp.ceptory.com/mcp. No key.

  2. 2

    Approve in your browser

    The agent opens Ceptory. Sign in if needed and click Allow, just like connecting Ceptory Desktop.

  3. 3

    Ask in plain language

    "Alert me when someone loiters at the gate for more than five minutes." The agent does the rest.

Setup

Add Ceptory to your client

Share one URL with your agent: https://mcp.ceptory.com/mcp. On first use it opens Ceptory in your browser to approve, so there is no key to copy. Agents without browser sign-in can use a token from the dashboard instead. Docs for LLMs: https://www.ceptory.com/mcp/llms.txt

Claude Code

Run in your terminal, then type /mcp in Claude Code and choose Authenticate. Your browser opens Ceptory to approve.

claude mcp add --transport http ceptory https://mcp.ceptory.com/mcp

Claude Desktop & claude.ai

Connect in one click

Or set it up by hand: Settings → Connectors → Add custom connector. Paste the URL, then click Connect and approve in the browser.

Name: Ceptory
URL:  https://mcp.ceptory.com/mcp

Or set it up by hand: Add to ~/.cursor/mcp.json (or .cursor/mcp.json in a project). Cursor asks you to sign in to Ceptory.

{
  "mcpServers": {
    "ceptory": {
      "url": "https://mcp.ceptory.com/mcp"
    }
  }
}

Or set it up by hand: Add to .vscode/mcp.json (GitHub Copilot agent mode), or run “MCP: Add Server”. VS Code asks you to sign in.

{
  "servers": {
    "ceptory": {
      "type": "http",
      "url": "https://mcp.ceptory.com/mcp"
    }
  }
}

Codex

Run in your terminal. Codex opens the browser to sign in.

codex mcp add ceptory --url https://mcp.ceptory.com/mcp
codex mcp login ceptory

Gemini CLI

Add to ~/.gemini/settings.json, then run /mcp auth ceptory in Gemini CLI.

{
  "mcpServers": {
    "ceptory": {
      "httpUrl": "https://mcp.ceptory.com/mcp"
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json. Windsurf asks you to sign in.

{
  "mcpServers": {
    "ceptory": {
      "serverUrl": "https://mcp.ceptory.com/mcp"
    }
  }
}

Any other agent

Any MCP client or agent framework (OpenAI Agents SDK, LangChain, Mastra, n8n…). Clients that support MCP auth discover sign-in from the URL. Otherwise send a token from the dashboard as a Bearer header, or run CEPTORY_TOKEN=<token> npx -y ceptory-mcp for stdio.

URL:        https://mcp.ceptory.com/mcp
Transport:  Streamable HTTP (MCP 2026-07-28, 2025 clients supported)
Auth:       OAuth 2.1 sign-in (discovered automatically), or Authorization: Bearer <token>

Examples

Things you can ask

  • “Set up a camera called Loading dock that alerts our Discord when anyone isn't wearing a helmet and a hi-vis vest.”
  • “Watch the back fence for people or vehicles between 10pm and 6am India time. Only the left half of the frame matters.”
  • “Which cameras are offline right now? Show me today's open alerts with their photos, then resolve the false alarms.”
  • “Raise the crowd limit on the lobby camera to 25 and pause the parking camera.”
  • “Put the gate, lobby and car park cameras in a group called North site and show me where people went in the last hour.”
  • “Make a chat called Night shift, add this video link, and tell me every time a white van appears.”
  • “Run the fire and smoke engine on this photo and tell me what it found.”

Reference

Tools

Each tool calls the same endpoint the dashboard uses, so permissions, limits and billing match the app exactly.

Cameras & workflows

list_camera_workflows
The 10 workflows (PPE, person down, fire & hazards, intrusion, face access, loitering, unattended bags, crowd limit, illegal parking, rider helmets) with every setting explained.
create_camera
Connect a camera with a workflow, settings, a watched zone and Discord alerts.
update_camera
Rename, change rules, schedule or zone, pause or resume.
list_cameras / get_camera
Status (online, offline, paused), settings, people and recent alerts.
add_camera_person / remove_camera_person
People lists for face access cameras: staff, watchlist, and your own allow or alert lists.
update_plate_list / read_plates
Allowed and alert-on-sight number plate lists for parking cameras; read plates from a photo.
geocode_address
Find an address to place a camera on the map.
send_camera_frame
Run a camera's rules on one frame and create any alerts.
get_camera_snapshot
The latest frame as an image.
test_discord_webhook
Check an alert channel works.
list_camera_groups / create_camera_group / update_camera_group / delete_camera_group
Sites and areas that group cameras.
get_movement / get_sighting_image
Journeys of people and vehicles across a group's cameras, with photos and map positions.
delete_camera
Remove a camera (agents ask first).

Alerts

list_alerts
Open, acknowledged or resolved alerts, by camera and time.
update_alert
Acknowledge, resolve or re-open.
get_alert_image
The annotated photo captured when the alert fired.

Video chats & search

list_chats / create_chat / get_chat / rename_chat / delete_chat
The same chats as the dashboard.
add_video_from_url
Index a video from an HTTPS link.
create_upload_url / complete_upload
Upload a local file (up to 2 GB). The npm bridge does this in one step with upload_video_file.
get_chat_videos / list_videos
Indexing status and every video in the workspace.
search_videos
Find moments in plain language, with start and end times.
ask_video / analyze_video / get_video
Answers, summaries and the indexed timeline of a video.
get_video_info / get_video_storyboard / get_video_frames
Watch a video yourself: a storyboard grid of timestamped frames, then full frames to zoom in. Free, no indexing wait.
retry_video / delete_video
Fix failed indexing, or remove a video.

AI engines & account

list_engines / run_engine
Run any single detector (PPE, fire, plates, faces, vehicles, crowd…) on an image, with the annotated result.
get_account
Who the agent acts as, the workspace, plan and prepaid balance.

Security

You stay in control

  • You approve each agent in your browser on ceptory.com. Agents never see your password, and only a hash of their session is stored.
  • Every connected agent is listed in the dashboard. Revoke one and it stops working at once.
  • Viewer roles stay read-only, and every call has the app's rate limits.
  • Delete tools are marked destructive, so clients ask you before running them.

FAQ

MCP questions, answered

What can an agent do through the Ceptory MCP server?

Everything you can do in the dashboard: create and configure cameras and workflows, manage staff lists and watchlists, read and resolve alerts with their photos, upload and search video, ask questions about a video, and run individual AI detection engines. It uses the same checks as the app, so viewers stay read-only.

Who does the agent act as?

You. When you approve the connection in your browser, the agent gets its own session for your account and workspace, like a Ceptory Desktop sign-in. Paid actions such as indexing, search and analysis use your workspace's prepaid balance.

How do I revoke access?

Open Connect AI agents in the dashboard and revoke the agent. It stops working at once. Browser sign-ins last 30 days, then the agent asks you to approve again.

Does the agent see my camera feeds live?

No. Cameras keep streaming from Ceptory Desktop or the browser. The agent sees what the dashboard sees: settings, status, the latest snapshot, alerts and their photos.

Does it only work with Claude?

No. It is a standard MCP server, so it works with any MCP client or agent: Claude Code and Claude Desktop, Cursor, VS Code, Codex, Gemini CLI, Windsurf, and agent frameworks such as the OpenAI Agents SDK or LangChain.

Which MCP version does it support?

MCP 2026-07-28 over Streamable HTTP, with fallback for clients on the 2025 protocol. The ceptory-mcp npm package bridges stdio-only clients such as Claude Desktop.

Get started

Connect your agent to Ceptory

Paste one URL, approve in your browser, and manage your cameras by talking to your agent.