Linkerd2 MCP

What is Linkerd2 MCP?

Linkerd2 MCP is a Model Context Protocol (MCP) server that enables you to interact with your Linkerd2 service mesh using natural language or API calls. It builds on Linkerd and Linkerd-viz to provide high‑level mesh introspection and safe policy mutation via a gRPC interface.

How to use Linkerd2 MCP?

Build the collector and MCP server binaries with go build -o bin/collector ./cmd/collector and go build -o bin/mcp-server ./cmd/mcp-server. Deploy on a local kind cluster using the provided Helm chart, then port‑forward the service and test the gRPC API with grpcurl. For example, grpcurl -plaintext localhost:10900 mcp.v1.MeshContext/GetMeshGraph retrieves the mesh graph, and grpcurl -plaintext -d '{"namespace":"default","name":"allow-foo","json_spec":"{\"foo\": \"bar\"}"}' localhost:10900 mcp.v1.MeshContext/ApplyAuthorizationPolicy applies an authorization policy.

Key features of Linkerd2 MCP

• gRPC MCP API for mesh introspection and policy mutation
• Lightweight, fault‑tolerant control plane for Linkerd service meshes
• In‑memory mesh model backed by Redis for snapshots and deltas
• Applies authorization policy updates directly to the mesh
• Built with the help of AI agents (active development)
• Helm chart for easy deployment on Kubernetes

Use cases of Linkerd2 MCP

• Fetch the current mesh graph to visualize service topology
• Apply Linkerd authorization policies programmatically or via chat
• Integrate mesh management into AI‑powered assistant workflows
• Automate policy changes across namespaces in a Linkerd cluster
• Introspect mesh health and connectivity without kubectl commands

FAQ from Linkerd2 MCP

What does Linkerd2 MCP do that the Linkerd CLI or viz cannot?

Linkerd2 MCP exposes a gRPC MCP API that allows external systems (including AI agents) to query and mutate mesh state in a structured, programmatic way, whereas the CLI and viz are primarily human‑facing tools.

What dependencies does Linkerd2 MCP require?

Go 1.22+, Docker or Podman, kubectl 1.30+, kind (for local development), Linkerd CLI 2.15+, Prometheus (from Linkerd viz), Buf CLI 1.29+, and Helm 3.14+.

Where does mesh data live?

Data is read from the Linkerd service mesh running in your Kubernetes cluster (Prometheus for metrics, Kubernetes API for resources) and cached in an in‑memory graph model with Redis snapshots.

What transport and authentication does the server use?

The MCP server exposes a gRPC API (plaintext by default). No authentication is described in the current documentation.

Is Linkerd2 MCP production‑ready?

No. The project is a Work In Progress (WIP) under active development. Features and APIs may change at any time.