NetClaw Techguide github.com/automateyournetwork/netclaw
comprehensive deep-dive · 12 sections

NetClaw: A CCIE-Level AI Network Engineering Agent

A technical guide to NetClaw — an autonomous, CCIE-level network-operations agent built on Anthropic Claude and the OpenClaw framework, orchestrating 191 skills across 113 MCP integrations with ITSM-gated change management and immutable audit trails.

SECTION 01

What NetClaw Is and the Problem It Solves

Introduces NetClaw as a CCIE-level autonomous network-engineering agent and frames the multivendor complexity problem it targets. Sets the stage for the rest of the guide.

Read section →
SECTION 02

The Layered Architecture at a Glance

Walks the five-layer stack from chat UI down through Claude/OpenClaw, the skills layer, the MCP server layer, and backend systems. The mental model everything else builds on.

Read section →
SECTION 03

The OpenClaw Foundation and Agent Workspace Context

Explains the OpenClaw runtime NetClaw is built on and the Markdown workspace-context files (SOUL, AGENTS, IDENTITY, USER, TOOLS, HEARTBEAT) that shape the agent's behavior each session.

Read section →
SECTION 04

The Skills Layer: 191 Domain Automation Routines

Details how skills package domain-specific network automation, the major skill categories (pyATS device, vendor-specific, observability, security, visualization), and skill-level gating.

Read section →
SECTION 05

The MCP Integration Layer: 113 Servers and Transports

Covers the Model Context Protocol servers that connect NetClaw to real infrastructure, the stdio-first transport choice, the domain coverage, and how skills invoke MCP tools.

Read section →
SECTION 06

Session Lifecycle and End-to-End Execution Flow

Traces a request from onboarding and setup through runtime context loading to a concrete task execution, showing how UI, agent, skill, MCP, and backend cooperate on a real query.

Read section →
SECTION 07

Change Management, ITSM Gating, and the GAIT Audit Trail

Explains the ServiceNow-gated change workflow (baseline → plan → apply → verify → document), automatic rollback, and the append-only Git-based GAIT audit that answers 'what did the AI do?'.

Read section →
SECTION 08

iN2N Federation: 'A Risk of NetClaws' and the NCFED Protocol

Describes the federated deployment model where specialized member claws are coordinated by a Border Claw, the token/least-privilege benefits, and the experimental NCFED inter-claw protocol.

Read section →
SECTION 09

Production Security Mode: DefenseClaw, Sandboxing, and Fail-Closed Design

Covers the layered security controls for production: systemd kernel confinement, the DefenseClaw guardrail proxy, honest posture reporting, and fail-closed behavior when controls are missing.

Read section →
SECTION 10

Token Optimization with GCF and Long-Term Memory with MemPalace

Explains the Graph Compact Format that compresses network data 55-83% to save tokens, and the MemPalace semantic/temporal memory that stores the 'why' behind changes separate from GAIT.

Read section →
SECTION 11

Dependencies, Installation, and Configuration

Enumerates the language runtimes, key libraries, and external services NetClaw depends on, then walks the interactive installer, install profiles, two-phase setup, and environment/config layout.

Read section →
SECTION 12

Notable Design Choices, Limitations, and When to Use It

Synthesizes the guiding design principles (stdio-first, fail-closed, append-only audit, least privilege) against the honest limitations (token explosion, ephemeral MCP state, experimental protocols) to guide adoption.

Read section →