MCP server

Cymo hosts the MCP server at:

https://mcp.cymatics.tech/mcp

Connect any MCP client that supports remote servers over Streamable HTTP. There's nothing to install or run. The server holds no data: each tool calls the API with your key, so the key's scopes, rate limits and metering apply, and a test key keeps every call on the simulator.

Send a Cymo API key from the dashboard as a bearer token: Authorization: Bearer cym_test_…. A request without one gets 401 missing_api_key. Use a cym_test_ key while you build and a cym_live_ key for your paired Cymos.

Replace cym_test_… with your key.

claude mcp add --transport http cymo https://mcp.cymatics.tech/mcp \
  --header "Authorization: Bearer cym_test_…"

{
  "mcpServers": {
    "cymo": {
      "url": "https://mcp.cymatics.tech/mcp",
      "headers": {
        "Authorization": "Bearer cym_test_…"
      }
    }
  }
}

{
  "mcp": {
    "servers": {
      "cymo": {
        "type": "http",
        "url": "https://mcp.cymatics.tech/mcp",
        "headers": {
          "Authorization": "Bearer cym_test_…"
        }
      }
    }
  }
}

export CYMO_API_KEY=cym_test_…
codex mcp add cymo --url https://mcp.cymatics.tech/mcp \
  --bearer-token-env-var CYMO_API_KEY

{
  "mcpServers": {
    "cymo": {
      "serverUrl": "https://mcp.cymatics.tech/mcp",
      "headers": {
        "Authorization": "Bearer cym_test_…"
      }
    }
  }
}

{
  "mcpServers": {
    "cymo": {
      "httpUrl": "https://mcp.cymatics.tech/mcp",
      "headers": {
        "Authorization": "Bearer cym_test_…"
      }
    }
  }
}

{
  "context_servers": {
    "cymo": {
      "url": "https://mcp.cymatics.tech/mcp",
      "headers": {
        "Authorization": "Bearer cym_test_…"
      }
    }
  }
}

Claude's custom connectors (Settings > Connectors) sign in with OAuth and can't send an API key, so in Claude Desktop use the local server. Any other client that can send a header on a remote server works with the same URL and header.

Add ?read_only=1 to the URL to list only tools that read, for an agent that should look but never change anything. The key's scopes still apply on top.

https://mcp.cymatics.tech/mcp?read_only=1

The hosted server runs the same code as @cymos/mcp. Run it yourself over stdio with your key in CYMO_API_KEY; add --read-only after the package name to list only read tools.

{
  "mcpServers": {
    "cymo": {
      "command": "npx",
      "args": [
        "-y",
        "@cymos/mcp"
      ],
      "env": {
        "CYMO_API_KEY": "cym_test_…"
      }
    }
  }
}

claude mcp add cymo -e CYMO_API_KEY=cym_test_… \
  -- npx -y @cymos/mcp

{
  "mcpServers": {
    "cymo": {
      "command": "npx",
      "args": [
        "-y",
        "@cymos/mcp"
      ],
      "env": {
        "CYMO_API_KEY": "cym_test_…"
      }
    }
  }
}

{
  "mcp": {
    "servers": {
      "cymo": {
        "type": "stdio",
        "command": "npx",
        "args": [
          "-y",
          "@cymos/mcp"
        ],
        "env": {
          "CYMO_API_KEY": "cym_test_…"
        }
      }
    }
  }
}

codex mcp add cymo --env CYMO_API_KEY=cym_test_… \
  -- npx -y @cymos/mcp

With cymatics mock start, the whole loop runs on your machine:

$ cymatics mock start --port 8787

# then run the MCP server against it
$ CYMO_API_URL=http://127.0.0.1:8787 CYMO_API_KEY=cym_test_local \
    npx -y @cymos/mcp

The package also serves Streamable HTTP at http://127.0.0.1:8788/mcp, with each client sending its own key as a bearer token, as with the hosted server.

$ npm install @cymos/mcp
$ PORT=8788 node node_modules/@cymos/mcp/dist/http.js

{
  "mcpServers": {
    "cymo": {
      "url": "http://127.0.0.1:8788/mcp",
      "headers": {
        "Authorization": "Bearer cym_test_…"
      }
    }
  }
}

The server has one tool per API operation, named after it and generated from the same contract, with the operation's input schema and read, write and destructive hints. Each API reference page names its tool. Tools for operations with a request body take it as input. Read-only mode hides the tools marked write.

• Project

  get_project

• Capabilities

  list_capabilities

• Models

  list_models get_model

• Sensing profiles

  list_sensing_profiles

• Retention profiles

  list_retention_profiles

• Coordinate frames

  list_coordinate_frames

• Skeleton definitions

  list_skeleton_definitions

• Rooms

  list_rooms create_room write get_room update_room write archive_room write get_zones replace_zones write get_calibration replace_calibration write get_processing_config replace_processing_config write stream_room

• Devices

  list_devices get_device update_device write get_device_health get_sensing_config replace_sensing_config write list_device_observations

• Scenes

  list_scenes create_scene write get_scene update_scene write delete_scene write

• Vitals

  get_vitals list_vitals_history

• Presence

  get_presence

• Movement

  get_movement

• Stress

  get_stress

• Skeleton

  get_skeleton

• Classifications

  get_classifications

• Results

  list_results

• People

  list_person_results stream_person

• Sleep

  list_sleep_sessions get_sleep_summary

• Activity

  get_activity_summary

• Analyses

  list_analyses get_analysis

• Events

  list_events

• Webhooks

  list_webhooks create_webhook write get_webhook update_webhook write delete_webhook write test_webhook write list_webhook_deliveries

• Scenarios

  list_scenarios create_scenario_run write

• Scenario runs

  get_scenario_run control_scenario_run write

• Raw data

  create_raw_data_export write get_raw_data_export

• Usage

  get_usage

• Request logs

  list_request_logs

Stream tools return a bounded window instead of an open stream. They watch for duration_ms (default 10 s, up to 60 s) or until max_samples messages (default 20, up to 200), then return them with truncated set when more may have followed. Raw radar bytes have no tool; use the raw-data export tools instead.

stream_room({ room_id: "room_bedroom", duration_ms: 30000, max_samples: 50 })

{
  "data": {
    "object": "stream_window",
    "operation_id": "stream-room",
    "sample_count": 50,
    "truncated": true,
    "ended_by": "max_samples",
    "messages": [{ "id": "…", "kind": "snapshot", "data": { … } }, …]
  },
  "meta": { "livemode": false, … }
}

• Run the bed_exit scenario and tell me when the person left bed.create_scenario_run, list_events
• Is anyone in room_bedroom right now? What are their heart rate and breathing?get_room, get_vitals
• Watch room_bedroom for 30 seconds and summarise what changed.stream_room
• How did the person in room_bedroom sleep last night?get_sleep_summary
• Send bed.exited and fall.detected events to my staging webhook.create_webhook, test_webhook
• Which of my requests failed today, and why?list_request_logs