Ask your plant questions in Claude or Cursor.
A plant-local MCP server that turns tags, assets, history and alarms into typed tools. Plug it into Claude, Cursor or your own agent and ask the factory questions in plain language.
Fourteen tools. One plant model.
Grouped into toolsets you can enable per client with ?toolsets=see,history. Every result is enriched with name, unit, quality and source.
Live values
Find tags by name or asset, read current values with unit and quality.
mii_see_find · mii_see_values
mii_see_describe · mii_see_readTime series
Ranges and last-known values from the plant historian — no export, no CSV.
mii_history_series
mii_history_lastAsset graph
contains · feeds · controls · depends_on. Ask "what stops if this stops?"
mii_understand_graph
mii_understand_impactPlant health
Inventory, device connection state, runtime status and active alarms.
mii_monitor_inventory · mii_monitor_devices
mii_monitor_runtimes · mii_monitor_alarmsProposals
The model can suggest a change. A person approves it in Console.
mii_act_proposalsIdentity
Which user, which role, which plant — so the model knows its limits.
mii_meta_whoamiAdd one server. Done.
Create an agent credential in Console, then export ready-made settings with ?export=claude or ?export=cursor. Streamable HTTP, JSON-RPC, nothing to install on the client.
# Claude Code claude mcp add --transport http mii \ https://plant.local/agent/v1/mcp \ --header "Authorization: Bearer mii_ac_••••" # Cursor · .cursor/mcp.json { "mcpServers": { "mii": { "url": "https://plant.local/agent/v1/mcp" } } }
Answers with a trail.
Because every value comes from a typed tool call, every answer can be traced back to a tag, a time range and a quality flag. No hallucinated readings.
- Shift handover summaries
- Root-cause from alarm floods
- "What changed since yesterday?"
- Engineering in Cursor with live plant context
tools/call mii_history_series { "tag": "Line2/P-101/vibration", "from": "-8h", "step": "5m" } → 96 points · unit mm/s · quality GOOD source mi-tsdb@plant-hub
Built for OT security teams, not around them.
Identity
Console RBAC
Each call carries a Console JWT or a scoped mii_ac_* credential. Roles filter which tools even appear.
Isolation
No PLC socket
The model talks to the hub, never to the floor. Field protocols stay inside Site Runtimes.
Change control
Propose, then approve
Writes become proposals with a human approval step. Every call and decision lands in the audit log.
Data residency
Plant-local
History never leaves the hub unless you send it. One plant, one historian.
Model choice
Bring any model
Claude over MCP, or a local OpenAI-compatible model for fully air-gapped sites.
Scoping
Toolsets per client
Give a shift assistant see,monitor and an engineer history,understand.
Ask your plant something real.
We'll connect one line and hand your team an MCP URL. Bring Claude or Cursor; we bring the context.