CHeKT Developers

What is MCP?

Model Context Protocol (MCP) is an open standard that lets LLM agents discover and call tools across services. An MCP server exposes a set of named tools with typed schemas; the agent reads the schemas, decides which tools to call, and chains them together to solve a task.

The @chekt/mcp package is the CHeKT MCP server. It wraps the Public API in MCP tool definitions tuned for LLMs — clear names, descriptive parameter docs, and structured responses that don't waste tokens.

How a prompt becomes an API call

When an operator types a prompt into Claude or ChatGPT, the agent breaks it down into discrete tool calls. The MCP server translates each call into authenticated HTTPS requests to CHeKT. Results flow back as structured JSON for the agent to reason over.

Operatornatural language

LLM agentClaude / ChatGPT

CHeKT MCP server@chekt/mcp

CHeKT APIapi.chekt.com

01
OperatorLLM agent
Plain-English prompt"Acknowledge motion alarms at warehouses"
02
LLM agentCHeKT MCP server
Tool callchekt.alarms.list(filter=...)
03
CHeKT MCP serverCHeKT API
Authenticated HTTPAuthorization: Bearer <app-key>
04
CHeKT APICHeKT MCP server
JSON responseAlarms matching filter
05
CHeKT MCP serverLLM agent
Tool resultStructured for LLM consumption
06
LLM agentOperator
Natural replyPlus follow-up tool calls as needed

Figure 1. The six-step path from a natural-language prompt to a CHeKT API call and back.

Set up MCP

You need three things: an MCP-capable client (Claude Desktop, ChatGPT Desktop, Cursor, or your own agent), a CHeKT App with an API key, and 60 seconds.

Claude Desktop

Open Claude Desktop's settings file and add the chekt server. Claude handles the rest on startup.

{
  "mcpServers": {
    "chekt": {
      "command": "npx",
      "args": ["-y", "@chekt/mcp"],
      "env": {
        "CHEKT_API_KEY": "your-app-api-key"
      }
    }
  }
}

Restart Claude and you'll see the CHeKT tools listed in the tools menu. Try a prompt: "What alarms fired in the last hour?"

Other MCP clients

Any client that speaks MCP works. The @chekt/mcp binary follows the standard stdio transport — point your client at it with the CHEKT_API_KEY env var, and you're done.

Available tools

Nine tools cover the most common operator and developer workflows. We add more as the MCP protocol matures and as partners request them.

chekt.alarms.list: List alarms with optional filters (status, site, priority).
chekt.alarms.get: Get full details for a specific alarm.
chekt.alarms.acknowledge: Acknowledge an alarm. Requires alarms:write.
chekt.alarms.dispatch: Dispatch a responder. Requires alarms:write.
chekt.devices.list: List devices with type and status filters.
chekt.devices.snapshot: Capture a fresh snapshot from a device.
chekt.sites.list: List sites for the current dealer.
chekt.partitions.arm: Arm a partition. Requires sites:write.
chekt.partitions.disarm: Disarm a partition. Requires sites:write.

A real session

Here's an annotated trace of a real Claude session combining multiple CHeKT MCP tools to handle a typical morning-shift triage task.

You: Show me all unacknowledged high-priority alarms at the East Bay sites
       today, and acknowledge any motion alarms with confidence > 80%.

Claude: I'll do that in a few steps.

  → chekt.sites.list(filter="East Bay")
    Found 3 sites: East Bay HQ, East Bay Warehouse, East Bay Lot.

  → chekt.alarms.list(
      sites=[3 IDs],
      status="new",
      priority="high",
      since="today",
    )
    Found 7 alarms.

  → chekt.alarms.get(id="A-10293")
    Trigger=motion. Confidence=0.92.

  → chekt.alarms.acknowledge(
      id="A-10293",
      reason="AI verified motion, confidence=0.92",
    )
    Acknowledged.

  ... (repeats for 4 more high-confidence motion alarms)

  Done. Acknowledged 5 of 7. The remaining 2 are panic-button events —
  I left them open for human review.

Safety & audit

Agents calling production monitoring systems demand discipline. We bake three guardrails into the MCP server so you don't have to.

Why MCP matters for CHeKT

The next wave of partners won't write SDKs — they'll write prompts. With MCP and the CLI, the CHeKT ecosystem is open to anyone who can describe what they want, even without traditional development skills. Small ops teams, individual security contractors, and AI-native businesses can all build on CHeKT.

At the same time, every action still runs through the same authenticated, audited surface. No special "AI mode" — just the same API with a thinner client on top.

Let agents work with CHeKT — natively.