Talk To Your Charger

Point an AI at your charger.

A cloud MCP server lets Claude, Cursor, and Codex read and control CoCan directly; agents such as OpenClaw can drive FluxAI® through MCP, while local Prometheus and JSON metrics keep the record. As far as we know, no other charger does this.

First Of Its Kind

As far as we know, no other charger ships an MCP server.

MCP — the Model Context Protocol — is the open standard for wiring AI tools into real systems. CoCan's cloud MCP server lets Claude, Cursor, Codex, or agents such as OpenClaw read live charger status, run a diagnostic, or turn intent into charging strategy.

Natural Language Control

Use plain language to read and schedule a charger.

"How many watts is port C3 pulling?" "Why is this cable stuck at 60W?" "Set the laptop port to high-performance." You can now just say these to an AI; over MCP it reads port voltage and current, PD negotiation, cable capability, and history, then suggests the next move or makes the change for you.

The MCP server organizes CoCan's capabilities into a set of tools: read device info, read live port details, read PD negotiation and cable data, pull historical power curves, and switch charging strategy, set thermal mode, allocate watts per port, or turn ports on and off. Your intent decides which tool the AI reaches for.

From SDC to AIDC, the change is that an external agent can understand a goal and translate it into port priority, thermal mode, and power scheduling. An agent such as OpenClaw calls Skills over MCP and uses Heartbeat to track device liveness and state; the actual power split is still executed by FluxAI®, the charging decision engine.

It also ships a handful of built-in prompts that bundle common tasks into one step: status brief, full status with PD detail, control actions, recording a charging session, comparing ports or before-and-after changes, and deep device-and-cable diagnostics.

A user working beside AI CoCan Mirror — AI Workflow **The charger becomes a schedulable node beside the computer.**
AI integration matters during real work: checking status, reading ports, and changing strategy all happen on the same desk.

MCP Surface

What the MCP server exposes

Read tools

Device info, live per-port V/A/W, charging status, thermal mode, PD negotiation and cable data, and historical power curves.

Control tools

Switch charging strategy, set thermal mode, allocate watts per port, and turn specific ports on or off.

Built-in prompts

Status brief, full status, control actions, session recording, port comparison, and device-and-cable diagnostics.

Standard transports

SSE and Streamable HTTP transports, ready for Claude Desktop, Cursor, Codex, and other clients.

Client Config

Connecting is one block of config

{
  "mcpServers": {
    "cocan": {
      "url": "https://<mcp-host>/<psn>/<token>/mcp"
    }
  }
}

Local Metrics

Want it fully local? Use Prometheus and JSON.

Not everyone wants to go through a cloud. CoCan's IonBridge firmware runs a local web server: one endpoint speaks Prometheus, another returns JSON. Point Prometheus and Grafana at it and you chart port power, temperature, and allocation on your own machines, with the data never leaving the local network.

Want a snapshot? One request to the JSON endpoint returns current ports, power, temperature, and status. Want history? The Prometheus endpoint is already in standard scrape format. A NAS, a Raspberry Pi, or a spare old laptop is enough to collect it and keep a long-running Grafana board on the charger.

What an agent did can be reviewed from the same data. Every allocation, mode switch, port toggle, and task result can land in MQTT telemetry, local Prometheus, or JSON; the safety boundary is still held by the chip-to-cloud trust chain, hardware headroom, and FluxAI®'s strategy constraints.

Three Paths In

Three paths in

Cloud MCP

Point Claude, Cursor, Codex, or OpenClaw at the cloud MCP server to read, diagnose, and schedule the charger in natural language.

Prometheus

The device serves standard Prometheus metrics locally for Prometheus and Grafana to scrape and chart over time.

JSON

One request returns current ports, power, temperature, and status for scripts and dashboards.

Open throughout

MQTT, protobuf, and datasheets are all public, so AI, scripts, and dashboards share one set of real data and one audit trail.

Get CoCan Mirror