Agentic AI Playground
Build, deploy, and test MCP servers and LangGraph AI agents
0/20 servers | 0/10 agents
LLM Provider Settings
Configure your LLM provider to power AI agents. Settings are saved in your browser and persist across sessions.
GCP Vertex AI
Access LLM via GCP (Anthropic models only)
Complete sign-in in the opened tab, then paste the verification code below.
Authenticated
Enter
global for global routing, or a specific region like us-east5 for data residency / region-scoped IAM.Enter your model name (e.g. claude-sonnet-4-6@default).
Paste Verification Code
Complete sign-in in the opened tab, then copy the verification code and paste it below.
MCP Servers
No MCP Servers
Deploy a mock server, connect to a remote MCP, or import an OpenAPI spec to get started.
Deploy Mock MCP Server
Relationships & Tools
Connect to Remote MCP Server
Import from OpenAPI Spec
Upload Your Own MCP Server Code
Paste your FastMCP Python server code below. The code must use FastMCP and read the port from os.environ.get("MCP_PORT").
Copy this configuration to your MCP client settings (Claude Code, Cursor, etc.) to connect to your running MCP servers.
Agents
No Agents Deployed
Use the Agent Designer tab to build and deploy an agent, or Upload Code to deploy your own.
Chat
Running
Mission Context
Mission Type
Active Mission
Remaining: --:--
Usage Attribution
Sent in _meta on every tools/call and tools/list
Client Context (key-value pairs passed in via _meta on every tools/call in this chat)
Send a message to start.
You are interacting with an AI system. Do not input personal or confidential information. Always independently review and validate any AI recommendations before implementing them.
Choose a template to preview and deploy an MCP server + Agent with pre-configured data and tools.
Design and deploy AI agents visually.
Configure an LLM, connect MCP tool servers, set agent instructions, and deploy — all from this canvas. Use View Code to inspect the generated agent source before deploying.
AgentMinder-Secured Agent Setup
1. Click Login to GCP Vertex in the top bar to authorize LLM access
2. Deploy MCP tool servers from the MCP Servers tab
3. Create an orchestrator client in AgentMinder and assign the AI Agent Orchestration Client role or another role with
4. Configure MCP servers as AI Resources in AgentMinder
5. Configure an AI Agent in AgentMinder and create access policies
6. Enable Secure with AgentMinder and enter your Base URL
7. (Optional) Provide an admin Access Token and click Apply to enable Find dropdowns for looking up orchestrator, agent, and MCP server details. Without a token, you can enter all details manually.
8. Click Deploy Agent
2. Deploy MCP tool servers from the MCP Servers tab
3. Create an orchestrator client in AgentMinder and assign the AI Agent Orchestration Client role or another role with
urn:iam:t.aiagentorchestrationclient permission4. Configure MCP servers as AI Resources in AgentMinder
5. Configure an AI Agent in AgentMinder and create access policies
6. Enable Secure with AgentMinder and enter your Base URL
7. (Optional) Provide an admin Access Token and click Apply to enable Find dropdowns for looking up orchestrator, agent, and MCP server details. Without a token, you can enter all details manually.
8. Click Deploy Agent
Find buttons will be enabled once a valid access token is applied.
Add AI Resource Server
LLM
Claude Opus 4.6
Changing the model will redeploy the agent
seconds (30–1800)
AI Agent
Redirect URI
0 connected
Orchestrator
minutes (10–60)
Upload Agent Code
Paste your LangGraph agent Python code below. The code must read the port from os.environ.get("AGENT_PORT") and expose /chat and /health HTTP endpoints.
Multi-Agent Workflows
No Multi-Agents Deployed
Use the Multi-Agent Designer tab to build and deploy a supervisor, or Upload Code to deploy your own.
Chat
Running
Mission Context
Mission Type
Active Mission
Remaining: --:--
Usage Attribution
Sent in _meta on every tools/call and tools/list
Client Context (key-value pairs passed in via _meta on every tools/call in this chat)
Send a message to start.
You are interacting with an AI system. Do not input personal or confidential information. Always independently review and validate any AI recommendations before implementing them.
Design and deploy multi-agent systems visually.
Configure an orchestrator, add sub-agents with MCP tool servers, and deploy a supervisor that coordinates them all. Use View Code to inspect the generated supervisor source before deploying.
One Level vs N-Level
Use the One Level / N-Level toggle to choose your agent hierarchy:
One Level — Supervisor delegates to a flat list of sub-agents, each with its own MCP tools. Suitable for most use cases.
N-Level — Sub-agents can have their own child agents, forming a multi-level hierarchy. Parent agents act as hybrid agents with both their own MCP tools and child agent-as-tool capabilities. With AgentMinder, each parent agent delegates tokens to its children for secure hierarchical access.
One Level — Supervisor delegates to a flat list of sub-agents, each with its own MCP tools. Suitable for most use cases.
N-Level — Sub-agents can have their own child agents, forming a multi-level hierarchy. Parent agents act as hybrid agents with both their own MCP tools and child agent-as-tool capabilities. With AgentMinder, each parent agent delegates tokens to its children for secure hierarchical access.
AgentMinder-Secured Multi-Agent Setup
1. Click Login to GCP Vertex in the top bar to authorize LLM access
2. Deploy MCP tool servers from the MCP Servers tab
3. Select One Level or N-Level mode depending on your agent hierarchy
4. Create an Orchestrator client in AgentMinder and assign the AI Agent Orchestration Client role with
5. Create an AI Agent app in AgentMinder for the Supervisor Agent
6. Create an AI Agent app in AgentMinder for each Sub-Agent (and each child agent in N-Level mode)
7. Register each MCP server as an AI Resource Server in AgentMinder and create access policies
8. Enable Secure with AgentMinder and enter your Base URL
9. (Optional) Provide an admin Access Token and click Apply to enable Find buttons on each card for looking up orchestrator, supervisor, sub-agent, and MCP server details. Without a token, you can enter all details manually.
10. Click Deploy Supervisor
2. Deploy MCP tool servers from the MCP Servers tab
3. Select One Level or N-Level mode depending on your agent hierarchy
4. Create an Orchestrator client in AgentMinder and assign the AI Agent Orchestration Client role with
urn:iam:t.aiagentorchestrationclient permission (this is a regular app, not an AI agent)5. Create an AI Agent app in AgentMinder for the Supervisor Agent
6. Create an AI Agent app in AgentMinder for each Sub-Agent (and each child agent in N-Level mode)
7. Register each MCP server as an AI Resource Server in AgentMinder and create access policies
8. Enable Secure with AgentMinder and enter your Base URL
9. (Optional) Provide an admin Access Token and click Apply to enable Find buttons on each card for looking up orchestrator, supervisor, sub-agent, and MCP server details. Without a token, you can enter all details manually.
10. Click Deploy Supervisor
Find buttons will be enabled once a valid access token is applied.
LLM
Changing the model will redeploy the supervisor
seconds (30–1800)
Orchestrator
minutes (10–60)
AI Supervisor Agent
Redirect URI
Upload Supervisor Code
Paste your LangGraph supervisor agent Python code below. The code must read the port from os.environ.get("AGENT_PORT") and expose /chat and /health HTTP endpoints.
Customer Support Demo
Not provisioned. Go to Configure Demos subtab → Customer Support Demo card to provision IDSP resources first.
Governed Agent
No vLLM selected
Checking...
No authentication required
Ungoverned Agent
Checking...
No authentication required
Status: Not started
Agents: Not deployed
Active Mission
JWT
Type:
Intents:
Expires:
Configure Demos
Provision/Teardown buttons will be enabled once a valid access token is validated.
HR Agent Demo
Not Provisioned
Purpose
Orchestrators: hr_agent_demo_orch
Agents: hr_agent_demo_agent
MCP Servers: hr-agent-demo (10 tools)
Resource Servers: hr_agent_demo_rs
Mission Types: hr_inquiry, hr_management
Intents: hr:read:dept, hr:read:employee, hr:manage:employee
Policies: 3 (read/manage by user group)
Auth Policy: hr_agent_demo_authn_policy (password, SMS OTP, FIDO or passkey)
Users: hra_admin, hra_user
User Password: AgentMinder@Demo1
Groups: hra_admins, hra_users
Security Config
Protected by Default: ON
Enforce Policies: ON
Mission Liveness: ON
Require Consent:
LLM Provider:
To use a Virtual LLM, configure it in the Agent Designer after provisioning
Use via HR Agent Demo chat UI in the Agents tab
Travel Agent Demo
Not Provisioned
Purpose
Orchestrators: travel_demo_orch
Agents: travel_demo_supervisor → travel_demo_flight_agent + travel_demo_hotel_agent
MCP Servers: travel-demo-flights (16 tools), travel-demo-hotels (16 tools)
Resource Servers: travel_demo_flight_rs, travel_demo_hotel_rs
Mission Types: travel_planning, flight_planning, hotel_planning
Intents: travel:read:flight, travel:book:flight, travel:read:hotel, travel:book:hotel
Policies: 6 (read/book by user group, plus deny-without-user per RS)
Auth Policy: travel_demo_authn_policy (password, SMS OTP, FIDO or passkey)
Users: tra_admin, tra_user
User Password: AgentMinder@Demo1
Groups: tra_admins, tra_users
Security Config
Protected by Default: ON
Enforce Policies: ON
Mission Liveness: ON
Require Consent:
Delegation Tags (supervisor → sub-agent)
flight_agent:
hotel_agent:
Comma-separated. Use in policy Context Conditions:
( ${delegation.tags} eq booking )
LLM Provider:
To use a Virtual LLM, configure it in the Agent Designer after provisioning
Use via Travel Agent Demo chat UI in the Multi Agents tab
Customer Support Demo
Not Provisioned
Purpose
Shows the difference between a governed and an ungoverned agent. Every prompt runs against two copies of the
same support agent side by side: the governed one is delegated by the signed-in
rep, bound to the customer on the call, and every tool call goes through the AgentMinder gateway; the
ungoverned one has the same model, the same prompt and the same tools against
its own copy of the backend, with no identity and no gateway. The customer's account notes carry a prompt
injection telling the agent to put the full card number in the refund confirmation email — the governed copy
is denied, because
crm_get_customer_payment_details
maps to payment_data_read, an intent
that is not in its auth surface; the ungoverned copy complies and leaks the card. A second guardrail holds
refunds over $100 until another rep approves, so the demo shows both what the gateway blocks outright and
what it escalates to a human.
Orchestrators: support_demo_orch
Agents: support_demo_agent (HYBRID, depth=1, elevated)
MCP Servers: support-demo-refund (9 tools, behind the gateway) + support-demo-refund-direct (ungoverned copy)
Resource Servers: support_demo_rs
Mission Types: resolve_billing_dispute
Intents: customer_lookup, review_read, payment_refund, customer_notify (4 of 8 in surface)
Policies: access grant (reps group) + refund guardrail (> $100 held) + call scope
Auth Policies: customers → password; reps → password + SMS OTP / FIDO / passkey
Users: reps dani.reyes, sarah.mitchell; customers robert.chen, lisa.park, james.wilson, maria.santos, david.kim
User Password: AgentMinder@Demo1
Groups: support_demo_reps, support_demo_customers
Security Config
Protected by Default: ON
Enforce Policies: ON
Mission Liveness: ON
Require Consent:
HITL Guardrails: payments_issue_refund > $100, approved by a second rep
LLM providers come from the Customer Support Demo tab: agents deploy at Provision; Start redeploys them only if the choice changed
Use via Customer Support Demo subtab
Progress Log
After provisioning, you may need to start agents from the Agents or Multi Agents tabs.
Waiting for action...
