Run these tools from Claude

Connector URL

https://www.futuremicromoldllc.com/api/mcp

Claude.ai and Claude Desktop

Settings → Connectors → Add custom connector, then paste the URL above. No account or key is needed.

Claude Code

claude mcp add --transport http fmm-lab https://www.futuremicromoldllc.com/api/mcp

Anything else that speaks MCP

The endpoint serves Streamable HTTP. For a client that only speaks stdio, bridge it with mcp-remote:

npx -y mcp-remote https://www.futuremicromoldllc.com/api/mcp

What Claude can do with it

Viscosity curve study viscosity_curve

Find the optimal injection velocity from a velocity / peak-pressure sweep. Computes relative viscosity (plastic pressure times fill time) against relative shear rate, locates the shear-thinning plateau, and picks a velocity from the middle of it. Record a fill time on every point for a result that transfers between machines. This is the first study in the sequence; every later study holds its velocity constant.

Cavity balance study cavity_balance

Quantify cavity-to-cavity fill imbalance from short-shot weights. Give two or more replicate shots at a fill level and the result also says whether the variation comes from the cavities (runner and gate geometry) or from shot-to-shot process drift — the two have opposite fixes. Judge balance on a short shot at 90% fill or above.

Pressure-drop study pressure_drop

Find where injection pressure is consumed along the flow path. Each reading is the CUMULATIVE peak pressure needed to fill through that section — an air shot gives the nozzle figure, then sprue, runner, gate and the full part — so the values ascend. Known sections: Nozzle, Sprue, Runner, Gate, Part. Run every reading at the velocity the viscosity curve selected.

Cosmetic process-window study cosmetic_window

Map which combinations of two process parameters produce a cosmetically acceptable part, and how much room surrounds the centre. Sweep wider than you expect the window to be: a passing point on the edge of the range tested means the window was never closed on that side. Axes and their units: hold_pressure (psi), melt_temp (°C), mold_temp (°C), injection_velocity (in/s), hold_time (s), cooling_time (s). Defects: short, flash, sink, burn, splay, weld-line, drag, warp.

Gate-seal study gate_seal

Find the hold time at which the gate freezes and part weight stops rising. Seal is the first of three consecutive weights within 0.1% of each other; the recommended production hold time adds a margin of one increment. A weight that falls back after seal means the gate did not stay frozen. Run at the velocity the viscosity curve selected.

Cooling-time study cooling_time

Find the shortest cooling time at which a dimension has stopped changing, and the cycle seconds that come off. Supply the drawing tolerance and the rule tightens to a tenth of it. Measure each part at a consistent interval after ejection — the part keeps moving.

Process capability (Cp, Cpk, Pp, Ppk) process_capability

Judge whether a running process can hold a dimension inside its tolerance. Reports Cp and Cpk from within-subgroup sigma and Pp and Ppk from overall sigma — the second pair is what a customer's SPC package computes from the same measurements. One specification limit is enough. Sample at least 25 consecutive parts without adjusting the process.

How to run a study get_study_protocol

Return the full procedure for one study: what it is for, the equipment needed, the step-by-step method at the press, and what a good result looks like. Call this before asking anyone for numbers, so the data that comes back is the data the calculation needs.

Plan the study sequence plan_process_development

Given the studies already run, return the full development sequence with each one marked done or outstanding, what each fixes, what it depends on, and which study to run next. The order matters: each study fixes a setpoint the next one holds constant, so running them out of order produces numbers that do not transfer.

Teach Claude the method too

The connector gives Claude the calculations. The skill below gives it the method — which study to run when, which columns to ask for at the press, and how to read a result that says its own confidence is low.

Download the skill

In claude.ai or Claude Desktop: Settings → Capabilities → Skills → upload the file.

Your data

A study is your process window, so it is treated as yours. The numbers you send are used to compute the result you asked for and are then discarded: they are not stored, not written to any database, and not sent to analytics. The share link the tools return carries your study in the URL fragment, which browsers never transmit to a server — the same mechanism the Copy-link button on every Lab page uses.

The calculations are the same ones this site runs in your browser. Nothing here asserts anything about any particular supplier; every number comes from the data you provide.