Connect AgentGuard To Claude Code

Use the current MCP Server or Skill setup to make AgentGuard available inside Claude Code, with npm and install.sh available for the local runtime.

View setup preview Read Docs

The public commands establish service access and local installation. They do not establish automatic interception of every Claude Code action.

Service connection pathPublic commands available

Claude CodeRequest-driven

MCP / SkillIntegration method

AgentGuardSecurity service

ConnectReviewEvidence

ConnectionMCP Server / Skill

Service behaviorRequest-driven

Delivery modelnpm / install.sh runtime

Current statusPublic commands available

01 / CONNECTION

How AgentGuard Connects

See which supported AgentGuard forms apply to this target and what connection work still remains.

MCP Server

Expose AgentGuard through Claude Code's native MCP tool interface using the current public command.

Agent Skill

Copy the AgentGuard skill into Claude Code so the agent can follow a named security-review workflow.

Local runtime

Install the AgentGuard command-line runtime through npm or the macOS and Linux install script.

Invocation boundary

Keep Claude Code actions that never call the MCP server, skill, or local runtime visible as uncovered behavior.

02 / CONNECT

Connect, Configure, Verify

Commands match the current public setup paths. Production documentation should add approved versions, prerequisites, rollback steps, and source verification.

Claude Code · current verified runtime commands

# Option 1: add AgentGuard as an MCP server
claude mcp add agentguard -- npx -y @anthropic-ai/mcp-remote@latest https://github.com/GoPlusSecurity/agentguard

# Option 2: install the AgentGuard skill
cp -r agentguard/skills/agentguard ~/.claude/skills/agentguard

# Local runtime options
npm install -g @goplus/agentguard
curl -fsSL https://agentguard.gopluslabs.io/install.sh | bash

Choose a connection

Use MCP Server for a native tool connection or Skill for an instruction-driven AgentGuard workflow.

Install what the path needs

Use the public command for the selected connection and add the local runtime through npm or install.sh when required.

Verify in Claude Code

Confirm discovery, invoke one controlled security request, and inspect the result returned to the agent.

03 / SURFACE

Integration Surface

These labels separate installation, service access, workflow connection, and runtime behavior.

MCP ServerPublic Claude command

SkillPublic Claude path

npm packageRuntime install

install.shRuntime bootstrap

Automatic interceptionNot established

VerificationRun in Claude Code

04 / EVIDENCE

Verify The Connection And Result

Check the connection, runtime, returned result, and coverage boundary before treating setup as verified.

Connection

Claude Code lists AgentGuard as an MCP server or discovers the installed skill.

Runtime

The npm or install.sh runtime passes the documented status and diagnostic checks when used.

Result

A controlled AgentGuard request returns reviewable output inside the Claude Code workflow.

Boundary

Actions that bypass every configured AgentGuard path remain outside the integration.

05 / LIMITS

Known Limitations Stay Visible

These statements prevent a documented integration from being read as universal protection.

Invocation

MCP tools and skills depend on the agent invoking the configured AgentGuard path.

Installation versus connection

npm and install.sh make the runtime available; they do not connect Claude Code by themselves.

Version support

Production documentation needs tested Claude Code, MCP client, skill, package, and installer versions.

Runtime claims

Do not claim universal blocking until the exact Claude Code action path is verified.

06 / FAQ

Integration Questions

Keep answers aligned with current product behavior and the last verified integration version.

Which Claude Code methods are publicly documented?+

The current homepage provides an MCP Server command and a Claude Code Skill copy command.

Why are npm and install.sh also shown?+

They install the local AgentGuard runtime that a supported workflow may use after connection.

Does setup automatically cover every Claude Code tool call?+

The current public setup evidence does not establish universal interception.

Claude Code integration

Verify The Integration Before Production

Run the documented setup, confirm AgentGuard is reachable from Claude Code, and test the behavior expected from this connection.

Open Quickstart Book a Demo