Skip to main content
Glama
giasonpooni

Atelier MCP

by giasonpooni
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

Atelier MCP

Observe-first Model Context Protocol server for small-format atelier mills. Token-gated rough and finish passes. Works with Claude, Grok, and Codex.

This is not a factory. It is the control plane a sculptor or maker can put in front of a machine they already own — foam, plaster, wood, soapstone, limestone — so an assistant can see the job and cannot fire the spindle unless a human returns a server-minted token bound to that exact pass.

Claude / Grok / Codex
        |  MCP stdio
        v
   atelier-mcp
        |  GuardLattice
        |  plane AND job AND units AND extract AND envelope
        |  AND magnitude AND live AND token_binding AND forbid
        v
   sim cell  (GRBL live path is opt-in, later)

There is no send_gcode. There is no request_jog. A finish pass ends in hand-off. The human finishes the work.

Status (0.1.0)

Plane

Tools

Observe

capabilities, cell_status

Prepare

bind_job, inspect_program, mark_handoff

Command

request_rough, request_finish_pass

Recover

feed_hold, emergency_stop

Default backend is an in-process simulator. Command is off until ATELIER_MCP_ALLOW_COMMAND=1.

Related MCP server: foundry net-industrial

Install

pip install git+https://github.com/giasonpooni/Atelier-MCP.git
atelier-mcp

From a clone:

pip install -e ".[dev]"
python -m pytest

Clients

Same binary. Configs in examples/.

Leave command and live off until you have read SAFETY.md.

Claude Desktop — examples/claude_desktop_config.json

Grok Build

grok mcp add atelier atelier-mcp

Codex — examples/codex_config.toml

A pass

1. bind_job(stock, bit, material=foam, extract_on=false)
2. inspect_program(gcode)          # static, not a proof
3. request_rough(...)              # preview + token
4. human reads the preview
5. request_rough(..., token=...)   # commit
6. request_finish_pass(...)        # same two-step
7. mark_handoff()                  # chisel is yours

Swap the target between step 3 and step 5 → token_payload_mismatch. Soapstone without extract_on=true → extract gate stays open. Finish stepdown 1.5 mm → magnitude stays open.

Environment

Variable

Default

ATELIER_MCP_ALLOW_COMMAND

0

ATELIER_MCP_LIVE

0

ATELIER_MCP_ALLOW_RECOVER

1

ATELIER_MCP_MAX_STEPDOWN_MM

2.0

ATELIER_MCP_MAX_FINISH_MM

0.4

What this is not

Not Monumental Labs. Not a 7-axis stone cell. Not pour-cell telemetry. Not a digital twin. Not a finished sculpture.

Optional CSE fields (element_id, world_digest) may be stored on a job; v0.1 does not yet refuse on disposition. See docs/ROADMAP.md.

License

Apache-2.0. No warranty. SAFETY.md.

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • SINT Protocol

    sint-ai
    A
    license
    A
    quality
    A
    maintenance
    Security-enforcing MCP proxy that sits between an AI agent and any number of downstream MCP servers, intercepting every tool call through a capability-token policy gateway that can allow, deny, or escalate to human approval before the call reaches any real tool. It also exposes built-in operator tools for approval workflows, audit trail queries, token management, voice/HUD output, and hierarchical
    21
    14
    Apache 2.0

  • Aura

    XRSec
    A
    license
    Not graded
    quality
    A
    maintenance
    Enables AI assistants to securely access local machine capabilities such as file operations, shell commands, Skills, and Pi Tools through a controlled MCP interface with token authorization and multiple tunnel options.
    19 npm
    1
    MIT