Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@operandi-mcp Identify the Samsung ME20H705MSS and get defrost procedure."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
mcp-name: cc.operandi/operandi
operandi_mcp_server.py exposes OPERANDI over the Model Context Protocol so any MCP-capable
agent host (Claude Desktop, Claude Code, or your own agent runtime) can identify an appliance and
pull its grounded, safety-checked, robot-executable operation package as native tools.
This is the "build for agents" surface: the customer is a robot's planner / an LLM agent, not a human reading PDFs.
Tool | What it does |
| Resolve observed nameplate text / panel labels / model → a catalog object (call first). |
| The robot-executable package: model-exact procedures, control map + grounding, per-step verification signals, recovery state machine, safety envelope. |
| Browse operable appliances (optionally by category). |
| Find appliances by function ( |
Related MCP server: Groundcrew
A general model, cold, gives confidently-wrong physical instructions on ordinary appliances a large
fraction of the time (OPERANDI Stage A: 24% of cold instructions were would-fail, including
invented buttons and cycles). Grounded in the package these tools return, that fell to 0
hallucinations, 96% exact. The tools turn "guess the buttons" into "read the manufacturer's ground
truth". See ../docs/BUSINESS_MODEL.md.
pip install operandi-mcpThat's it. No key needed for your first packages: on first use the server mints an
instant trial key itself (POST /v1/trial — no signup, 2 operation packages included,
cached at ~/.operandi/mcp_key). When the trial is spent, tool responses tell the agent
exactly how to sign up free (10 packages/month) or go Pro.
Have a key already? Set it and it wins over the trial:
export OPERANDI_API_KEY=ok_live_...{
"mcpServers": {
"operandi": { "command": "operandi-mcp" }
}
}(Optionally add "env": {"OPERANDI_API_KEY": "ok_live_..."} once you have an account key.)
Then ask the agent: "Identify the Samsung ME20H705MSS and give me the safe procedure to defrost
0.5 kg of mince." — it will call identify_appliance then get_operation_package and answer from
grounded data.
The server is a thin, API-key-authenticated REST client — the same binary works against local dev
or the hosted service by changing OPERANDI_API_URL.
Auth is Authorization: Bearer <key>; a missing/invalid key surfaces as a tool error, not a crash.
Transport is newline-delimited JSON-RPC 2.0 (the MCP stdio transport). Offline protocol tests:
pytest tests/test_mcp_server.py.
Find appliances by what they DO (e.g. 'heat', 'wash', 'brew', 'defrost') — useful when an agent has a goal but not a specific model in mind.
Inputs: capability (required).
Get the robot-executable operation package for an appliance: model-exact ordered procedures, a control map with grounding (where each control is + how it is actuated), per-step verification signals, a recovery state machine, and a safety envelope (hazards / interlocks / never-do). This is the grounded knowledge a general model hallucinates without. Optionally target one procedure (e.g. 'defrost').
Inputs: ref (required), procedure.
Resolve which appliance an agent/robot is looking at, from observed text or panel labels, to an OPERANDI catalog object. Call this first, then get_operation_package with the returned slug. Returns ranked matches with confidence.
Inputs: query (required), brand, limit, category, +1 more.
Browse the appliances OPERANDI can operate, optionally filtered by category.
Inputs: limit, category.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 4 tools
Each tool has a clearly distinct purpose: find by capability, identify from observation, get operation package, and list all appliances. There is no overlap; the descriptions clearly differentiate them.
All names use snake_case and are descriptive, but 'find_by_capability' breaks the verb_noun pattern used by the others (identify, get, list). This is a minor inconsistency.
With 4 tools, the server covers the essential workflow (browse, search, identify, retrieve details) for its niche domain. It is slightly minimal but well-scoped, so score 4.
The set covers browsing, searching by capability, identification from observation, and retrieving detailed operation packages. Missing features like direct model lookup or procedure listing are minor gaps.
Codebase intelligence for agents: 152 structured artifacts across 21 programs, one call.
Evidence-grounded, graph-connected, correctable memory for agents.
Certified SEC EDGAR fact memory for AI agents with zero hallucination and filing provenance.
Verifiable Earth ground truth for AI agents: water, hazard, ground stability, resource, with proof.
