SSM (State Space Models) — MCP Tool Specification
Version: v1.1.9 | Status: Active | Last Updated: March 2026
Overview
This document defines the MCP tools exposed by the ssm 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.
Auto-Discovery
| Property | Value |
|---|---|
| Discovery method | @mcp_tool decorator scan |
| Namespace | ssm |
| Trust default | Safe |
| PAI bridge | src/codomyrmex/agents/pai/mcp/ |
Tool Reference
ssm_forward
Description: Run a forward pass through Mamba State Space Model. Trust Level: Safe Category: analysis
Parameters:
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
sequence_length | int | No | 8 | Sequence length to process |
d_model | int | No | 16 | Model dimension |
d_state | int | No | 8 | SSM state dimension |
n_layers | int | No | 2 | Number of Mamba blocks to stack |
Returns: dict — Dictionary with output_shape, d_model, d_state, n_layers, and status.
Example:
from codomyrmex.ssm.mcp_tools import ssm_forward
result = ssm_forward(sequence_length=16, d_model=32, d_state=16, n_layers=4)
Notes: Requires numpy. Generates random input data internally for the forward pass.
flash_attention_forward
Description: Run Flash Attention and verify against standard attention. Trust Level: Safe Category: analysis
Parameters:
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
seq_len | int | No | 16 | Sequence length |
d_model | int | No | 32 | Q/K/V dimension |
block_size | int | No | 8 | Flash attention tile size |
Returns: dict — Dictionary with output_shape, max_error_vs_standard (should be < 1e-5), passed (bool), and status.
Example:
from codomyrmex.ssm.mcp_tools import flash_attention_forward
result = flash_attention_forward(seq_len=32, d_model=64, block_size=16)
Notes: Imports from codomyrmex.neural.flash_attention. Generates random Q/K/V matrices internally. The passed field indicates whether the flash attention output matches standard attention within tolerance.
Integration Notes
- Auto-discovered: Yes (via
@mcp_toolinmcp_tools.py) - Trust Gateway: All tools are safe — no trust check required
- PAI Phases: VERIFY (model correctness validation), BUILD (neural architecture experimentation)
- Dependencies:
numpy,codomyrmex.neural.flash_attention(for flash_attention_forward)