Keycloak MCP Server
A complete Model Context Protocol (MCP) server for Keycloak 26.x
Keycloak MCP Server
A comprehensive Model Context Protocol (MCP) server that exposes the Keycloak Admin REST API as typed MCP tools. 299 tools covering all API categories.
Features
- Complete API coverage: All 299 Keycloak Admin REST API endpoints
- Dual transport: stdio (Claude Code) and SSE (GitHub Copilot, other MCP clients)
- Auto-authentication: Supports both password and client credentials flows with automatic token refresh
- Zero configuration tools: Each tool is self-describing with full input schemas
API Categories
| Category | Tools |
|---|---|
| Attack Detection | 3 |
| Authentication Management | 38 |
| Client Certificates | 6 |
| Client Initial Access | 3 |
| Client Registration Policy | 1 |
| Client Role Mappings | 10 |
| Client Scopes | 10 |
| Clients | 33 |
| Components | 6 |
| Groups | 11 |
| Identity Providers | 15 |
| Keys | 2 |
| Organizations | 13 |
| Protocol Mappers | 14 |
| Realms Admin | 37 |
| Roles | 28 |
| Roles by ID | 10 |
| Scope Mappings | 29 |
| Users | 30 |
| Total | 299 |
Demo
(Add a GIF or screenshot here demonstrating 3 real prompts and their executed tools!)
Installation
Option 1: Run directly with uvx (Recommended)
You can run the server directly without manual installation using astral's uv:
uvx keycloak-mcp-server
(When using uvx, you can pass environment variables inline or keep them in your MCP config file.)
Option 2: Install via pip
If you prefer a global or virtual environment installation:
pip install git+https://github.com/paoloamato2/keycloak-mcp-server.git
Option 3: Install from source (For development)
git clone https://github.com/paoloamato2/keycloak-mcp-server.git
cd keycloak-mcp-server
uv pip install -e .
Configuration
Set environment variables (or create a .env file based on .env.example):
# Required
export KEYCLOAK_URL=http://localhost:8080
# Authentication - Option A: Password flow
export KEYCLOAK_ADMIN_USERNAME=admin
export KEYCLOAK_ADMIN_PASSWORD=admin
# Authentication - Option B: Client credentials flow
export KEYCLOAK_CLIENT_ID=my-client
export KEYCLOAK_CLIENT_SECRET=my-secret
# Optional
export KEYCLOAK_ADMIN_REALM=master # default: master
export KEYCLOAK_VERIFY_SSL=true # default: true
Usage
Claude Code (stdio)
Add to your Claude Code MCP configuration (~/.claude/claude_desktop_config.json or project-level):
{
"mcpServers": {
"keycloak": {
"command": "python",
"args": ["-m", "keycloak_mcp_server"],
"env": {
"KEYCLOAK_URL": "http://localhost:8080",
"KEYCLOAK_ADMIN_USERNAME": "admin",
"KEYCLOAK_ADMIN_PASSWORD": "admin"
}
}
}
}
Or if installed with uv:
{
"mcpServers": {
"keycloak": {
"command": "uv",
"args": ["run", "--directory", "/path/to/keycloak-mcp-server", "python", "-m", "keycloak_mcp_server"],
"env": {
"KEYCLOAK_URL": "http://localhost:8080",
"KEYCLOAK_ADMIN_USERNAME": "admin",
"KEYCLOAK_ADMIN_PASSWORD": "admin"
}
}
}
}
GitHub Copilot (SSE)
Start the server with SSE transport:
python -m keycloak_mcp_server --transport sse --port 8080
Then configure in your GitHub Copilot MCP settings (VS Code settings.json):
{
"github.copilot.chat.mcpServers": {
"keycloak": {
"type": "sse",
"url": "http://localhost:8080/sse"
}
}
}
Command Line
# stdio mode (default)
python -m keycloak_mcp_server
# SSE mode
python -m keycloak_mcp_server --transport sse --host 0.0.0.0 --port 8080
# Using the entry point
keycloak-mcp-server --transport sse --port 8080
Security & Production Recommendations
⚠️ SECURITY WARNING: This MCP Server registers all Keycloak Admin REST API endpoints (299 tools), including sensitive write operations (like creating/deleting users, resetting passwords, and managing realms). Do not use your master realm super-admin credentials in a production environment.
When attaching this MCP server to your AI Assistants, please strictly follow the Principle of Least Privilege:
-
Use Service Accounts (Client Credentials Flow): Avoid using the Password flow (
KEYCLOAK_ADMIN_USERNAME/KEYCLOAK_ADMIN_PASSWORD). Instead, create a dedicated Keycloak Client with Service Accounts Enabled, and use theKEYCLOAK_CLIENT_IDandKEYCLOAK_CLIENT_SECRET. -
Limit Target Realms: Do not attach the server to the
masterrealm unless specifically necessary. PointKEYCLOAK_ADMIN_REALMto the exact realm your AI assistant should manage. -
Grant Only Required Roles: Only assign the minimum necessary roles to your MCP Service Account.
- If your LLM only needs to read data: Assign only
view-users,view-clients, orview-realm. - If your LLM needs to manage users: Assign only
manage-users. - Never assign
adminorrealm-adminroles to the AI unless you are fully aware of the risks.
- If your LLM only needs to read data: Assign only
-
Always Verify SSL: Keep
KEYCLOAK_VERIFY_SSL=trueenabled in production to prevent Man-in-the-Middle (MITM) attacks. Setting it tofalseis only acceptable for local development.
Examples
Once connected, you can use natural language to interact with Keycloak:
- "List all realms" → calls
list_realms - "Create a user called john in the master realm" → calls
create_user - "Show me all clients in the production realm" → calls
list_clients - "What roles does user X have?" → calls
get_user_role_mappings - "Add the admin role to the developers group" → calls
add_group_realm_role_mappings
Project Structure
src/keycloak_mcp_server/
├── __init__.py # Package entry point
├── __main__.py # CLI entry point
├── config.py # Environment-based configuration
├── client.py # Async HTTP client with auto-auth
├── server.py # MCP server setup and tool registration
└── endpoints/ # Endpoint definitions by category
├── __init__.py # Base classes (EndpointDef, Param)
├── attack_detection.py
├── authentication.py
├── certificates.py
├── client_initial_access.py
├── client_registration_policy.py
├── client_role_mappings.py
├── client_scopes.py
├── clients.py
├── component.py
├── groups.py
├── identity_providers.py
├── key.py
├── organizations.py
├── protocol_mappers.py
├── realms.py
├── roles.py
├── roles_by_id.py
├── scope_mappings.py
└── users.py
License
MIT
相關伺服器
Top GUN — GEO-Lens Audit API
Audits brand visibility across Large Language Models — returns a score, citations, and GEO recommendations.
MCP Simple Timeserver
Provides current local time, UTC time, and timezone information.
Kalshi MCP Server
A simple MCP server to interact with prediction market Kalshi
Proxenio MCP Server
MCP server for the Proxenio verified intent network enabling AI agents to retrieve verified matches and accept introductions with inherited trust controls.
timeService-mcp
A simple Model Context Protocol (MCP) server that provides the current date and time.
maxia-marketplace
AI-to-AI marketplace on 14 blockchains. 46 MCP tools for crypto swap (71 tokens), tokenized stocks (25), GPU rental, DeFi yields, LLM fine-tuning, wallet analysis. Pay in USDC.
Universal Image MCP
Universal MCP server for AI image generation supporting AWS Bedrock (Nova Canvas), OpenAI (GPT Image, DALL-E), and Google Gemini (Imagen 4). Generate, transform, and edit images using multiple AI models through a single Model Context Protocol interface.
mcp-server-openai-bridge
Bridge to OpenAI API. Access GPT, GPT-o and other OpenAI models through MCP.
DICOM MCP Server
Enables AI assistants to query, read, and move data on DICOM servers like PACS and VNA.
Mnemo Cortex
Persistent cross-agent semantic memory for AI agents. Recall past sessions, share knowledge across agents. Multi-agent (isolated writes, shared reads), local-first (SQLite + FTS5), works with any LLM — local Ollama at $0 or cloud APIs like Gemini and OpenAI. Integrations for Claude Code, Claude Desktop, and OpenClaw.