The Operating System for Agentic Intelligence
mAIndala is the trust & governance control plane for agentic AI. We maintain a verified catalog of MCP services, Skills, and Agents — then give organizations one place to govern the tools, credentials, and actions of every agent they run, including agents built on other platforms.
Why We Are Called mAIndala
A mandala (Sanskrit for “circle”) is a geometric design from Hindu, Buddhist, and other contemplative traditions: a unified center surrounded by concentric, symmetrical layers. Our name places AI at the center of the mandala — the full story. That is not decoration — it is the architecture.
mAIndala is pronounced “main-dah-la.” The name is mandala with AI woven into its center — m·AI·ndala — so the capital letters simply mark the acronym that lives inside the word. It is an enterprise trust and governance platform for autonomous AI agents.
- One center, not another tool. Organizations adopting agents do not need another framework — they need one place where every agent, whoever built it, is inventoried, risk-scanned, and governed. mAIndala is built to be that center.
- Order that radiates outward. In a mandala, structure tightens toward the center. So does ours: approval before an agent is registered, verification before a capability is installed, and policy enforced at the moment of every call — one audit trail behind all three.
- Symmetry — every agent, the same rules. An agent built on LangGraph, CrewAI, Copilot Studio, Bedrock, in-house, or on mAIndala itself lands in the same inventory, under the same policy, behind the same kill-switch. The outer rings will keep growing — more agents, more tools, more providers — and a mandala does not lose coherence as its rings expand, because everything stays oriented to the center.
- The control plane — Inventory · Verify · Control · Broker & revoke
- Your capability estate — Skills · MCP services · Connections
- Your agents, wherever built — LangGraph · CrewAI · Copilot Studio · Bedrock · in-house · mAIndala
Agents Don’t Just Answer — They Act
AI agents connect to tools, hold credentials, and take real actions. As adoption grows, capabilities spread across teams and platforms — LangGraph, CrewAI, Copilot Studio, Bedrock, and in-house apps — and most run outside any central oversight. Unverified components, unclear permissions, and shadow agent workflows become operational and security risk.
Agents built on LangGraph, CrewAI, Copilot Studio, Bedrock, or in-house don’t have to live outside your control. mAIndala inventories them, risk-scans their definitions, and routes their tool and data access through a policy gateway with a full audit trail and an instant kill-switch.
One control plane for every agent
mAIndala governs agents at the boundaries every agent has to cross — no matter where the agent was built: LangGraph, CrewAI, Copilot Studio, Bedrock, in-house, or on mAIndala itself.
Inventory
Build a single registry of every agent operating in your organization, each with an Agent Bill of Materials: what it is, what it can touch, and how risky it is.
Verify
Risk-scan agent and capability definitions with OWASP-aligned checks, approve or block, and get alerted when an approved agent later changes.
Control
Route agent tool and data access through a policy gateway with allow/deny rules, rate limits, time-of-day windows, and DLP redaction — every governed call is recorded.
Broker & revoke
Agents reach tools with scoped, short-lived credentials issued from an encrypted vault. Revoke access instantly with a one-click kill-switch.
What We Offer
MCP Services
3,400+ MCP servers indexed, spanning every domain — searchable by capability, transport, auth method, and Trust status. Queryable by AI agents at inference time via our MCP Gateway.
Skills
Reusable prompt packages that encode how to perform a specific task. Install a Skill to unlock its prompts in Claude, your IDE, or any agent harness. Create, verify, and publish your own. 1,800+ Skills and growing.
Agents
Multi-step agent blueprints that chain Skills, MCP tools, and built-in capabilities into complete workflows. Build manually or use the AI Agent Builder. Deploy to a live cloud endpoint or run locally with one command.
Private Groups
Curate a vetted catalog of Skills, Agents, and MCP services for your team. Members see only what you have approved. Group API keys give agents scoped access to the approved stack with no individual install required.
Cloud Deployment
Deploy to Google Cloud Run, AWS ECS Fargate, Azure Container Apps, or IBM Code Engine. Bring your own cloud credentials and LLM API key.
What Makes mAIndala Different
mAIndala governs an agent at four boundaries — its definition, the credentials it acts with, the tools it reaches, and the record it leaves — wherever the agent was built.
What an agent is
Every agent gets a single inventory entry and an Agent Bill of Materials, risk-scanned with OWASP-aligned checks before approval and re-checked whenever its definition changes.
What it can act with
Tool access uses scoped, short-lived credentials issued from an encrypted vault, so a long-lived secret is never handed to the agent for that purpose.
What it can reach
Tool and data access routed through a policy gateway is checked against allow/deny rules, rate limits, time-of-day windows, and DLP redaction — and every governed call is recorded.
What it did
A full audit trail and a one-click kill-switch, independent of which platform the agent was built on.
mAIndala Verified
Skills and Agents are scanned before they are published. Every catalog item carries a trust status — Verified, Partial, or Unverified — with the date it was last scanned, so you can tell a current result from a stale one.
Private Catalog for Your Team
Create Skills and Agents visible only to your group, add private MCP server URLs, and control who can publish to the broader catalog. A built-in publish flywheel takes content from group-private → unlisted → public, with governance gates at each step.
Bring Your Own LLM
Deploy to Google Cloud Run, AWS ECS Fargate, Azure Container Apps, or IBM Code Engine. Use Anthropic, OpenAI, Google Gemini, xAI, or your own OpenAI-compatible endpoint. mAIndala does not lock you into a specific model provider or cloud — bring your credentials, we provide the runtime.
AI Agent Builder
Describe what you want an agent to do in plain language. The platform searches the real catalog for matching Skills and MCP services and assembles a complete deployable agent — nodes, routing logic, and execution configuration included. No manual wiring required.
Run Locally, Free
Any agent in the catalog runs on your laptop with one command: npx maindala run agent/<slug>. The agent starts in Docker with your own LLM API key. No cloud account, no subscription. Explore and test before you commit to cloud deployment.
Our Mission
The shift from AI assistants to AI agents is not incremental — it is architectural. Agents need to act: browse the web, call APIs, read and write files, coordinate with other agents, and do so reliably across thousands of services. The bottleneck is no longer model capability. It is infrastructure.
As agents move from experimentation to production, a second challenge emerges: trust. Enterprises and individuals alike need to know that the Skills they install, the Agents they deploy, and the MCP services they connect carry no hidden risks. mAIndala addresses this directly: Skills and Agents are scanned before publication, every catalog item carries a trust status, and the whole catalog is governed by a transparent publish process.
mAIndala exists to be the governed center of that world: one control plane where every agent an organization runs — built here or anywhere else — is inventoried, verified, and controlled. And beneath it, the foundation that makes governance practical: a universal catalog of MCP-compatible services, a library of verified Skills, an expanding roster of deployable Agents, and the infrastructure to create, curate, and deploy all of them — from a private team environment to a live cloud endpoint.
How mAIndala Helps
mAIndala is the governed center where agents and the services they need meet. We provide:
- One control plane — A single registry, OWASP-aligned verification, call-time policy enforcement, and brokered short-lived credentials for every agent your organization runs — wherever it was built. One audit trail and a one-click kill-switch behind it all.
- Universal discovery — A searchable, filterable catalog — 3,400+ MCP servers indexed — queryable by capability, category, auth method, transport, and rating.
- mAIndala Verified — Skills and Agents are scanned before they are published. Every catalog item carries a trust status — Verified, Partial, or Unverified — with the date it was last scanned, so you can tell a current result from a stale one.
- Skills library — A growing collection of community-authored and mAIndala-verified prompt packages that encode domain expertise. Install a Skill to unlock its prompts in any AI environment.
- Agents catalog — Ready-to-deploy multi-step agent blueprints. Each Agent bundles Skills, routing logic, and cloud configuration — deploy to a live endpoint or run locally in Docker with a single CLI command.
- Private Catalog — Groups and enterprises curate vetted, private collections of Skills, Agents, and MCP services. One Group API key gives agents scoped access to the approved stack — no individual installs required.
- Cloud Deployment — Deploy any Agent to a live HTTPS endpoint on Google Cloud Run, AWS ECS Fargate, Azure Container Apps, or IBM Code Engine. Bring your own cloud and LLM credentials — mAIndala provides the runtime, orchestration, and activity monitoring.
- Agent-native API — An MCP Gateway agents can connect to at inference time — discover services, install Skills, and deploy Agents programmatically without human intervention.
- Open platform — Any service provider can list their MCP server. Any developer can publish a Skill or Agent. The catalog auto-indexes from public registries and surfaces new entries immediately.
Data Sources & Attribution
The mAIndala catalog is built on open data and community-contributed listings. Auto-indexed entries are sourced from the following open registries, used in accordance with their respective licenses:
Community-curated list of MCP servers, maintained by Frank Fiegel.
Official reference implementations and registry maintained by the MCP steering group under LF Projects.
Community-submitted services are provided by their respective owners. If you are a service owner and would like to update or remove your listing, please use the community forum or contact us directly.
Govern, curate, and deploy — on your terms
Whether you are governing agents across your organization, curating a private team catalog, or building and deploying agents yourself — mAIndala is your starting point.