Skip to content

Security

MCP Mesh provides three layers of security for production agent deployments:

flowchart TB
    L3["<b>Layer 3: Authorization</b><br/>WHO can do WHAT<br/>Header propagation + application-layer auth"]
    L2["<b>Layer 2: Agent-to-Agent mTLS</b><br/>Every inter-agent call is mutually authenticated"]
    L1["<b>Layer 1: Registration Trust</b><br/>Trust-entity verification before joining the mesh"]

    L3 --- L2 --- L1

    style L3 fill:#7ba7cc,color:#fff,stroke:#6b97bc
    style L2 fill:#6baa8e,color:#fff,stroke:#5b9a7e
    style L1 fill:#cca76b,color:#fff,stroke:#bc975b

Security is opt-in — local development works with no TLS by default. You can incrementally adopt stricter modes as you move toward production.

TLS Modes

Mode Description Use Case
off No TLS, plain HTTP Local development
auto Registry verifies certs if presented, allows without Transitional rollout
strict mTLS required — registry rejects connections without valid certificates Production

Quick Start

# Zero-config TLS (auto-generates CA + certs)
meshctl start --registry-only --tls-auto -d
meshctl start my_agent.py --tls-auto

meshctl start --registry-only -d \
  --env MCP_MESH_TLS_MODE=strict \
  --env MCP_MESH_TLS_CERT=/etc/certs/registry.pem \
  --env MCP_MESH_TLS_KEY=/etc/certs/registry-key.pem \
  --env MCP_MESH_TRUST_BACKEND=filestore \
  --env MCP_MESH_TRUST_DIR=/etc/certs/trust

meshctl start my_agent.py \
  --env MCP_MESH_TLS_MODE=strict \
  --env MCP_MESH_TLS_PROVIDER=vault \
  --env MCP_MESH_VAULT_ADDR=https://vault:8200 \
  --env MCP_MESH_VAULT_PKI_PATH=pki_int/issue/mesh-agent \
  --env VAULT_TOKEN=s.xxxxx

Put the Vault CA that issues agent certificates in /etc/certs/trust on the registry, as a .pem file. The filestore backend trusts only the .pem CAs in that directory (and its entities/ subdirectory), so without it every agent certificate is rejected.

meshctl start my_agent.py \
  --env MCP_MESH_TLS_MODE=strict \
  --env MCP_MESH_TLS_PROVIDER=spire \
  --env MCP_MESH_SPIRE_SOCKET=/run/spire/agent/sockets/agent.sock

Sections

  • Registration Trust

Registry verifies an agent's trust entity before accepting registration. Supports file-based certs, Vault PKI, and SPIRE workload identity.

Registration Trust

  • Agent-to-Agent mTLS

Every inter-agent call is mutually authenticated with TLS certificates. Works across Python, TypeScript, Java, and Go.

Agent-to-Agent mTLS

  • Authorization

Control which agents and users can access capabilities using header propagation and application-layer auth frameworks.

Authorization