Softmax Optimization — MCP Tool Specification
Version: v1.1.9 | Status: Active | Last Updated: March 2026
Overview
This document defines the MCP tools exposed by the softmax_opt module.
These tools are auto-discovered by the PAI MCP bridge via the @mcp_tool decorator
in mcp_tools.py and surfaced as part of the ~303 dynamic tools available to Claude.
The softmax optimization module provides numerically stable softmax variants including standard, log-softmax, and online softmax implementations.
Auto-Discovery
| Property | Value |
|---|---|
| Discovery method | @mcp_tool decorator scan |
| Namespace | softmax_opt |
| Trust default | Safe |
| PAI bridge | src/codomyrmex/agents/pai/mcp/ |
Tool Reference
compute_softmax
Description: Compute softmax probabilities from logits with support for multiple variants. Trust Level: Safe Category: analysis
Parameters:
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
logits | list[float] | Yes | -- | Raw unnormalized scores |
temperature | float | No | 1.0 | Temperature scaling (above 1 is more uniform and below 1 is more peaked) |
variant | str | No | "standard" | Softmax variant: "standard", "log", or "online" |
Returns: dict — Dictionary with status, probabilities (list), log_probs (list), entropy (float), max_prob (float), and sum_check (float, should be ~1.0).
Example:
from codomyrmex.softmax_opt.mcp_tools import compute_softmax
result = compute_softmax(logits=[1.0, 2.0, 3.0], temperature=0.5, variant="standard")
Notes: Temperature scaling only applies to the "standard" variant. The "online" variant uses a numerically stable single-pass algorithm. The "log" variant computes log-softmax directly for better numerical precision.
Integration Notes
- Auto-discovered: Yes (via
@mcp_toolinmcp_tools.py) - Trust Gateway: All tools are safe — no trust check required
- PAI Phases: BUILD (neural network components), VERIFY (numerical stability checks)
- Dependencies:
numpy,softmax_opt.kernel(softmax, log_softmax, online_softmax)