3D打印农场管理MCP服务器

odin-print-farm-mcp

Model Context Protocol server for O.D.I.N. — drive an actual print farm with AI agents, or run the bundled reference calculators standalone.

v2 (2026-04-15) splits the surface in two:

• Live tools (22) — agent-driven farm operation. Require a running O.D.I.N. instance + a scoped token. Queue jobs, pause printers, manage inventory, clear alerts, log maintenance, read the dashboard.
• Reference tools (4) — standalone calculators. No ODIN deployment required. Print-cost math, printer recommendations, farm capacity planning, software comparison.

Both surfaces ship in the same npm package; the live tools error at invocation if ODIN_BASE_URL and credentials aren't set, so the reference tools keep working for pre-sales and research use cases.

Quick Start

Standalone (reference tools only)

npx -y odin-print-farm-mcp@2

Wire into Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "odin-reference": {
      "command": "npx",
      "args": ["-y", "odin-print-farm-mcp@2"]
    }
  }
}

Agents can now use reference.calculate_print_cost, reference.recommend_printer_for_farm, reference.estimate_farm_capacity, reference.compare_farm_software.

Live (connected to O.D.I.N.)

1. Mint an API token in O.D.I.N. (Settings → API Tokens → New Token) with scope agent:read or agent:write.
2. Export the env vars and wire the client:

{
  "mcpServers": {
    "odin": {
      "command": "npx",
      "args": ["-y", "odin-print-farm-mcp@2"],
      "env": {
        "ODIN_BASE_URL": "http://192.168.1.100:8000",
        "ODIN_API_KEY": "odin_xxxxxxxxxxxxxxxxxxxx"
      }
    }
  }
}

All 26 tools are now available.

Tools

Live reads (scope agent:read or higher)

Live writes (scope agent:write or admin)

Every write tool accepts optional dry_run and idempotency_key.

Reference (no backend)

Agent Primitives

The live surface inherits four retry-safety primitives from the O.D.I.N. v1.8.9 backend:

Idempotency-Key. Every write tool auto-generates a UUID key per call; pass a stable idempotency_key across retries to deduplicate. On replay, the response comes back with _idempotent_replay: true.

X-Dry-Run. Pass dry_run: true to any write tool. The backend returns {dry_run: true, would_execute: {...}} without committing. Per-route opt-in — individual routes land preview branches in subsequent releases.

Structured errors. Failures return an error: {code, detail, retriable} envelope. Stable codes agents can branch on:

next_actions hints. Write responses include next_actions: [{tool, args?, reason?}] suggesting follow-up calls. Pure hint — no enforcement. Designed for 7B–32B local models.

ITAR / CMMC Deployment

O.D.I.N. ships ODIN_ITAR_MODE=1 for fail-closed air-gap deployments. Typical stack:

   Ollama (Qwen2.5-32B or similar)
      │
      ▼
   MCP client (Claude Desktop / OpenClaw / etc.)
      │ stdio
      ▼
   odin-print-farm-mcp@2  ── ODIN_BASE_URL=http://localhost:8000
      │ HTTP
      ▼
   O.D.I.N. backend  ── ODIN_ITAR_MODE=1
      │ LAN
      ▼
   Printers (LAN-only)

Zero outbound packets. All tokens, prompts, and telemetry stay inside the compliance boundary. See the ITAR / CMMC mode docs.

Migration from v1

v2.0.1 renames the four v1 tools into a reference.* namespace:

The behavior of each tool is unchanged — only the identifier. Pin odin-print-farm-mcp@1 if you need the un-namespaced IDs. Upgrade to @2 to access the 22 live tools plus the namespaced reference calculators.

Skill packages

The skills/ directory ships operator-ready prompt packs for Claude Code, OpenClaw, Cline, and other MCP clients. odin-farm/SKILL.md is the baseline "safe farm operator" rulebook — dry-run-first, branch on error codes, never batch cancellations. Drop it into ~/.claude/skills/ (Claude Code) or your client's skills directory, and a local 32B model will stop trying to cancel jobs without confirmation.

See skills/README.md for install + customization.

Development

git clone https://github.com/HughKantsime/odin-mcp
cd odin-mcp
npm install
npm run build
npm test  # 22 integration tests against an in-process mock ODIN

Tests use a node-native http mock server (test/mock_server.ts) — no external fixtures, no ODIN instance required, runs in ~1s.

Links

• O.D.I.N. — the backend this talks to
• MCP docs on runsodin.com
• Model Context Protocol spec
• npm package

License

Apache 2.0.

ODIN 2.0.13 authentication and permissions

Set ODIN_BEARER_TOKEN to an access JWT for Bearer authentication instead of ODIN_API_KEY. When both are explicitly configured, both headers are sent; ODIN rejects an invalid explicit API key at its perimeter even with a valid JWT. The dry-run version probe uses the same credentials. No login, token renewal or credential fallback is performed by this server.

Scoped user tokens require agent:read for reads or agent:write for writes (and reads); admin scope is also accepted. The owning user's role and the current Settings → Permissions action matrix still apply. JWT sessions bypass token-scope checks but retain role/action checks. Legacy global API keys resolve to ODIN's backend admin; they are not scoped user tokens. License and Education workflow restrictions remain server-enforced; scopes do not override them.

Printer state and spool material/availability filters run locally because ODIN does not accept those query parameters. Orders use status_filter; job states include scheduled and rejected, and order states include partial and fulfilled. Queue and summary calls page through all job/alert results. These multiple reads are not an atomic snapshot of a changing farm.

See the compatibility audit for every tool's route, permissions, fixes and follow-up capabilities.