MAF project lifecycle
What to do first, and what comes next, when you implement a MuleSoft Agent Fabric project. Follow the phases in order the first time; afterwards the lifecycle repeats as you refine the network.
-
Phase 1. Learn and prepare
Understand the building blocks and confirm you have the access you need before touching the platform.
Steps
- Learn how LLMs, RAG, MCP, and A2A fit together.
- Get your data foundations AI-ready (clean, governed, discoverable).
- Confirm Anypoint Platform permissions for Runtime Manager, Anypoint Code Builder, API Manager, and Exchange.
- Install Anypoint Code Builder and allowlist its URLs; plan your target (CloudHub 2.0 or Runtime Fabric).
Lessons
- LLM mechanics and prompt fundamentals AI-Native Skills
- Model Context Protocol (MCP) Essentials AI-Native Skills
- Agent2Agent (A2A) Protocol Essentials AI-Native Skills
- Building AI-ready data ecosystems Data Foundations
Official docs
-
Phase 2. Discover existing assets
Find the agents, MCP servers, and APIs you already have so you reuse them instead of rebuilding.
Steps
- Review your portfolio and Anypoint Exchange for existing agents, MCP servers, and APIs.
- Add scanners to keep the inventory current.
- Decide what to reuse, wrap, or build new.
Lessons
Official docs
-
Phase 3. Enable agent actions
Make the APIs and agents you already run available as tools. Add a broker only when work spans specialized agents.
Steps
- Expose existing APIs as MCP servers.
- Use an A2A bridge for agents that are not A2A-compliant.
- Bind MCP tools through connectors.
Lessons
- Model Context Protocol (MCP) Essentials AI-Native Skills
- Agent2Agent (A2A) Protocol Essentials AI-Native Skills
- Connectors and MCP tool binding Runtime & Connectors
Official docs
-
Phase 4. Set up the environment
Step 1 of the agent network procedure: prepare where the network will run and the gateway that fronts it.
Steps
- Choose a CloudHub 2.0 shared or private space, or a Runtime Fabric deployment target.
- Associate a private space with the business groups and environments you will deploy to.
- Set up Managed Omni Gateway for the space or target (a single gateway for ingress and egress is the recommended configuration).
Choose your target carefully
After you deploy an agent network you cannot redeploy it to a different target or gateway. To switch, first delete the existing Runtime Manager application, then redeploy.Lessons
- Deploying agents on CloudHub 2.0 Runtime & Connectors
- Managing network topographies and egress control Platform Engineering
- Automating infrastructure (IaC) Platform Engineering
Official docs
-
Phase 5. Build the agent network
Steps 2 to 4 of the procedure: create the project, define the network, and publish its assets.
Steps
- Create an agent network project with MuleSoft Vibes, Anypoint Code Builder, or the Anypoint CLI.
- Define the network in agent-network.yaml and exchange.json (brokers, agents, MCP servers, and their links).
- Publish the project to Exchange, which creates an asset for each broker, agent, and MCP server.
Lessons
- Designing autonomous workflows UAPIM & Agentic Enterprise
- APIs as an AI control plane UAPIM & Agentic Enterprise
Official docs
-
Phase 6. Govern and control
Put Omni Gateway in the request path so authentication, policy, and cost controls apply to live traffic.
Steps
- Apply policies and governance strategies once, across agents, APIs, and MCP servers.
- Secure A2A and MCP interactions (authentication, access control, PII handling).
- Cap and track LLM spend with model proxies and rate limiting.
Lessons
- Govern with Omni Gateway policies UAPIM & Agentic Enterprise
- Cost management and governance strategies Platform Engineering
Official docs
-
Phase 7. Deploy
Step 5 of the procedure: run the network instance on your target so governed traffic starts flowing.
Steps
- Deploy the agent network instance with MuleSoft Vibes, Anypoint Code Builder, or the Anypoint CLI.
- Confirm the policies you applied now enforce real traffic through the gateway.
Lessons
Official docs
-
Phase 8. Observe
See how agents, brokers, and MCP servers connect, and watch latency, errors, and volume in production.
Steps
- Visualize the network in Agent Visualizer.
- Monitor metrics, logs, and traces in Anypoint Monitoring.
- Configure alerts so regressions and rogue behavior surface quickly.
Lessons
Official docs
-
Phase 9. Assure with testing (forward-looking)
Turn what you observe into repeatable tests so each change to the network is verified before and after release.
Steps
- #
- Isolate agents with A2A mocks and follow the multi-broker test pyramid.
- Add protocol conformance, graph, auth, and governance tests, then automate them.
Forward-looking
This phase is a recommended practice from this learning hub, not a step in the official Agent Fabric procedure. Confirm the details against current MuleSoft documentation.Lessons
- The agent testing problem and layered evaluation Testing & Assurance
- Golden datasets, trace-to-test, and drift detection Testing & Assurance
- The multi-broker test pyramid Testing & Assurance
- Protocol conformance and spec validation Testing & Assurance
- Graph-level, auth, and governance testing Testing & Assurance
- Automating tests with gateway policies and Platform APIs Testing & Assurance
Official docs
- #
It is a loop, not a line