Skip to content

From Dev to Production

MCP Mesh covers every phase of the agent lifecycle — from first scaffold to production observability.

graph LR
    L["Learn"] --> D["Develop"] --> T["Test"] --> Dp["Deploy"] --> S["Secure"] --> O["Observe"]
    O -.-> D

    style L fill:#7ba7cc,color:#fff,stroke:#6b97bc
    style D fill:#6baa8e,color:#fff,stroke:#5b9a7e
    style T fill:#8eb8d4,color:#fff,stroke:#7ea8c4
    style Dp fill:#6baa8e,color:#fff,stroke:#5b9a7e
    style S fill:#cca76b,color:#fff,stroke:#bc975b
    style O fill:#7ba7cc,color:#fff,stroke:#6b97bc

Learn

meshctl --help              # All available commands
meshctl man                 # Built-in documentation for every feature
meshctl man llm             # Deep dive into any topic
meshctl man llm --java      # Language-specific variants

Documentation: mcp-mesh.ai | CLI: meshctl man <topic> | YouTube: youtube.com/@MCPMesh

Develop

meshctl scaffold --name my-agent --agent-type tool        # Generate agent
meshctl scaffold --name my-llm --agent-type llm-agent     # Generate LLM agent
meshctl scaffold --name my-api --agent-type llm-provider  # Generate provider

meshctl start --registry-only -d                          # Start registry
meshctl start my-agent -d                                 # Start agent (detached)
meshctl start my-agent --watch                            # Auto-restart on changes

meshctl list                                              # See registered agents
meshctl list --tools                                      # See available tools
meshctl status                                            # Agent health and dependencies
meshctl call greet --params '{"name": "World"}'           # Call a tool
meshctl logs my-agent                                     # View agent logs

Guide: Local Development | Reference: meshctl CLI

Test

meshctl call my-tool --params '{"input": "test"}'         # Call tools directly
meshctl call my-tool --trace                              # Call with tracing
meshctl trace <trace-id>                                  # View call tree

Zero-config mocking. Need to test against a payment service that isn't ready yet? Start a mock agent locally with the same capability name — the mesh wires it in automatically. No mock frameworks inside your code, no test configurations to manage, no risk of mock code reaching production. Your mock is just another agent on the mesh, and since it's not in your CI/CD pipeline, it can't accidentally ship. Compare this to traditional mocking (Spring @MockBean, Python unittest.mock) where mock code lives alongside production code and a misconfiguration silently executes the wrong path.

The TSuite integration testing framework runs 300+ tests across Python, TypeScript, and Java using this approach.

Deploy

meshctl scaffold --compose                    # Generate docker-compose.yml
meshctl scaffold --compose --observability    # With Grafana/Tempo/Redis
docker compose up -d

helm install mcp-core oci://ghcr.io/dhyansraj/mcp-mesh/mcp-mesh-core \
  --version 1.0.0 -n mcp-mesh --create-namespace

helm install my-agent oci://ghcr.io/dhyansraj/mcp-mesh/mcp-mesh-agent \
  --version 1.0.0 -n mcp-mesh -f helm-values.yaml

Same agent code runs locally, in Docker, and in Kubernetes — no changes needed.

Guide: Deployment Patterns | Kubernetes

Secure

# Development (zero-config TLS)
meshctl start --registry-only --tls-auto -d
meshctl start my-agent --tls-auto

# Production (Vault/SPIRE)
meshctl start my-agent \
  --env MCP_MESH_TLS_MODE=strict \
  --env MCP_MESH_TLS_PROVIDER=vault \
  --env MCP_MESH_VAULT_ADDR=https://vault:8200

Three layers: registration trust verifies identity, agent-to-agent mTLS authenticates every call, header propagation enables fine-grained authorization.

Observe

meshctl call my-tool --trace                    # Trace a call
meshctl trace <trace-id>                        # View distributed trace tree

Grafana dashboards and Tempo distributed tracing come pre-configured with --observability. Every tool call, every LLM hop, every inter-agent request — traced end-to-end.

Guide: Observability