菜单
GNN Model Context Protocol processing and tool registration. Use when registering GNN operations as MCP tools, building MCP server configurations, or integrating GNN capabilities with LLM tool-use workflows.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Generalized Notation Notation (GNN) processing pipeline for Active Inference generative models. Use when working with GNN files, running the 25-step pipeline, parsing model specifications, generating simulations, or producing visualizations and reports from GNN model definitions.
GNN command-line interface dispatch and health checks. Use when invoking the `gnn` CLI, validating subcommand availability, or checking command-line access to pipeline modules.
GNN LLM-enhanced analysis and model interpretation. Use when generating natural language descriptions of GNN models, getting AI-assisted model explanations, or performing LLM-powered analysis of Active Inference specifications.
GNN Language Server Protocol integration. Use when starting editor diagnostics, checking hover behavior, or validating LSP support for GNN files.
GNN pipeline orchestration and configuration management. Use when configuring pipeline execution, setting step inclusion/exclusion, managing pipeline state, or customizing the 25-step execution flow.
GNN code generation for simulation frameworks. Use when generating PyMDP, RxInfer.jl, ActiveInference.jl, JAX, DisCoPy, PyTorch, NumPyro, Stan, or bnlearn code from GNN model specifications.
| name | gnn-mcp-protocol |
| description | GNN Model Context Protocol processing and tool registration. Use when registering GNN operations as MCP tools, building MCP server configurations, or integrating GNN capabilities with LLM tool-use workflows. |
Processes Model Context Protocol (MCP) configurations and registers GNN pipeline operations as MCP tools. Enables LLM agents to invoke GNN capabilities through standardized tool interfaces.
# Run MCP processing
python src/21_mcp.py --target-dir input/gnn_files --output-dir output --verbose
# As part of pipeline
python src/main.py --only-steps 21 --verbose
from mcp import (
MCP, MCPRegistry, MCPServer, JSONRPCServer, MCPTool, MCPResource,
create_mcp_server, start_mcp_server,
register_tools, register_module_tools,
get_available_tools, list_available_tools,
list_available_resources, process_mcp,
get_mcp_instance, handle_mcp_request,
initialize,
)
# Process MCP step (used by pipeline)
process_mcp(
target_dir, output_dir,
verbose=True,
performance_mode="high",
strict_validation=True,
cache_ttl=120.0,
)
# Create and start a JSON-RPC server bound to the global MCP registry
server = create_mcp_server() # returns JSONRPCServer
start_mcp_server()
# Discover and register all pipeline modules' tools on the global MCP
register_tools() # populates mcp_instance
register_module_tools("gnn") # named module only
# Query available tools and resources
tools = get_available_tools() # same as list_available_tools()
tools_md = list_available_tools()
resources = list_available_resources()
MCP — core registry (tools, resources, modules, performance metrics).MCPRegistry — alias of MCP, the core registry.MCPServer — JSON-RPC 2.0 request handler bound to an MCP instance.JSONRPCServer — JSON-RPC 2.0 request handler bound to an MCP instance.MCPTool / MCPResource — dataclasses backing the registry.create_mcp_server / start_mcp_server — construct and start a JSON-RPC server.register_tools — discover and register all pipeline modules' tools on the global MCP.register_module_tools — register a single named module.initialize — check SDK + discover modules; exposes performance_mode,
enable_caching, enable_rate_limiting, strict_validation, cache_ttl,
modules_allowlist, per_module_timeout, overall_timeout, force_refresh.process_mcp — pipeline entry point (forwards overrides to initialize).get_available_tools, list_available_tools, list_available_resources.MCPToolNotFoundError, MCPValidationError, MCPModuleLoadError,
MCPInvalidParamsError, MCPToolExecutionError, MCPSDKNotFoundError,
MCPResourceNotFoundError, MCPPerformanceError.The MCP singleton honours these knobs, all propagated through
initialize() and process_mcp(**kwargs):
| Knob | Type | Default | Purpose |
|---|---|---|---|
performance_mode | "low"/"high" | "low" | Bulk toggle for caching, rate limiting, strict validation |
enable_caching | bool | from mode | Enable result cache |
enable_rate_limiting | bool | from mode | Enable per-tool rate limiting |
strict_validation | bool | from mode | Enforce JSON-schema validation of params |
cache_ttl | float (sec) | 300.0 | Result-cache TTL |
modules_allowlist | list[str] | None | Only load these modules under src/ |
per_module_timeout | float | 30.0 | Max seconds per module during discovery |
overall_timeout | float | 120.0 | Wall-clock budget for parallel discovery |
force_refresh | bool | False | Re-discover modules even if already loaded |
output/21_mcp_output/This module registers tools with the GNN MCP server (see mcp.py):