MCP Server

LangGuard exposes a Model Context Protocol (MCP) server so AI agents can connect directly to your governance control plane — either to self-enforce your policies, or to operate LangGuard on your behalf.

Navigation: Settings → MCP Server (/settings/mcp)

Because policies are served live, updates propagate instantly: change a policy in LangGuard and connected agents pick up the new guardrails on their next run — no agent restarts or config drift.

Two ways to use it

Agent compliance guardrails (read scope)

Give coding agents, AI assistants, and automated workflows read-only access so they operate within your governance. Agents can:

• Query active policies as machine-readable guardrails (get_rules)
• Check for violations after taking actions
• Browse the entity catalog (agents, tools, models)
• Search and inspect traces for context
• Read dashboard stats and analytics

Use a read-scope API key — agents can query but cannot modify anything.

Security & IT operations (write scope)

Let a security or IT-ops agent actively manage LangGuard. In addition to everything above, write-scope agents can:

• Create and update governance policies
• Acknowledge and resolve violations
• Tag entities with metadata and classifications
• Toggle policies on/off and change enforcement mode

Use a write-scope API key.

Connecting an agent

The MCP Server settings page gives you everything you need:

1. MCP endpoint — https://<your-instance>/api/mcp (HTTP transport / WebMCP). Browser-based and remote clients connect here.
2. API key — pick a key whose scope matches your use case (read for compliance, write for operations).
3. Configuration snippet — copy the ready-made snippet for your runtime. Tabs are provided for Claude Code, Cursor, VS Code, and HTTP / WebMCP.

A generic HTTP MCP configuration looks like:

{
  "mcpServers": {
    "langguard": {
      "type": "http",
      "url": "https://<your-instance>/api/mcp"
    }
  }
}

Copy the exact snippet from the settings page — it embeds your selected API key.

Quick setup for Claude Code

Install the LangGuard plugin from the marketplace:

/plugin marketplace add LangGuard-AI/langguard-plugins
/plugin install langguard@langguard-plugins

This installs the governance skills, a SessionStart hook, an auto-connected MCP server, and stores your API key in the OS keychain. You'll be prompted for:

• api_endpoint — your instance URL
• api_key — a key from this tenant
• governance_mode — compliance (read-only enforcer) or operations (read-write agent); match this to your API-key scope

Restart your session after install to pick up the SessionStart hook.

How agent compliance works

For the compliance use case, LangGuard turns your policies into machine-readable guardrails:

1. Agent starts a session and calls get_rules to fetch all active policies as compact guardrails.
2. Agent works within the guardrails, checking for violations after actions.
3. An operator updates a policy in LangGuard — no agent-side config change.
4. Agents auto-update — the next run receives the updated guardrails automatically.

Available tools

Read-scope tools are available to every connected agent; write-scope tools require a write key.

Tool	Scope	Purpose
get_rules	read	Fetch active policies as compact, agent-friendly guardrails
list_policies	read	List policies with 24h violation/scan stats
list_violations	read	List violations with status/severity filters
list_entities	read	Browse agents, tools, models, and identities
list_traces / get_trace	read	Search and inspect agent traces
get_dashboard_stats / get_analytics	read	Dashboard stats and analytics
create_policy / update_policy	write	Author and update governance policies
update_violation_status	write	Acknowledge or resolve violations
add_entity_tag / remove_entity_tag	write	Tag entities with metadata/classifications
toggle_policy	write	Enable/disable a policy or change its mode

…plus additional tools (24+ in total). The settings page lists the full set.

Next steps

• API Keys — create the read- or write-scope key to connect with
• Creating Policies — the guardrails agents enforce
• Trace Explorer — what agents can query