Solace Broker MCP Server: Focused Tools for AI-Assisted Broker Operations

The Solace Broker MCP Server lets any Model Context Protocol (MCP)-compatible AI assistant work with your Solace event brokers through purpose-built tools. Instead of handing an AI the full management API, which fills the context window, increases token costs, and adds load to production event brokers, we focused on critical functionality our customers need for broker management.

Solace Broker MCP Server Demo

Solace Product Manager showing off some of the capabilities using Claude Code and Solace Agent Mesh

This includes 47 named tools: 29 read-only tools for looking at things, and 18 write tools for changing them. The write tools are switched off until you opt in to avoid accidental changes in production. It is open source (Apache 2.0), works with software event brokers and appliances, and runs as a standalone binary, a Docker image, or on Kubernetes.

What It Supports

  • Broker health and status. Redundancy, replication, and overall status for any configured broker. Ask “What’s prod-broker’s current status?” and get a readable answer instead of nested Solace Element Management Protocol (SEMP) JSON.
  • Queues, clients, and Message VPNs. Message VPNs, queues, clients, client usernames, and client profiles. Ask “Which clients are connected to this Message VPN?” and get an answer without building the query yourself.
  • Troubleshooting. Queue metrics, slow subscribers, and discard diagnostics. Ask “Why is orders.q backing up?” and the assistant checks queue metrics and slow subscribers for you.
  • Integration points. REST Delivery Points, bridges, and Kafka Receivers and Kafka Senders. Ask “Are any bridges down?” without opening three separate screens.
  • Multiple brokers from one server. Each broker is addressed by alias. Ask “Compare message rates between prod-broker and dev-broker” and one question covers both.
  • Provisioning, when you allow it. Create, update, and delete Message VPNs, queues, queue subscriptions, topic endpoints, and REST Delivery Points, plus delete queued messages, disconnect clients, and clear statistics. Ask “Create a queue orders.q and subscribe it to orders/>” and it runs the two steps in order.
  • Schema introspection. The assistant can look up the fields that the management API accepts before it creates or changes anything.
  • Solace Agent Mesh. Wire the server into an Agent Mesh project as an MCP-tooled agent, so your agents can reason about the brokers underneath them.

Why This Beats Opening Up the Full API

Pointing an AI at the full management API means it can see, and potentially call, everything your credential can. It can also bog down the context window and cause hallucinations by mixing up the calls.  The MCP server narrows that down in ways an API key cannot.

  • A small, named toolbox. The assistant picks tools like get-queue-metrics and list-slow-subscribers rather than working out which of hundreds of API paths to call. That means fewer incorrect calls, and answers come back in a form the assistant can use.
  • Read-only by default. Write tools need an explicit opt-in. Your first rollout can be read-only.
  • Your identity, not a shared one. With OAuth 2.1 and OpenID Connect (OIDC), validated with Keycloak and Microsoft Entra ID, the caller’s identity is passed through to the broker when the broker is configured for OAuth. Broker audit logs show who asked, not just “the service account.”
  • Permissions that follow your groups. Individual tools can be gated by a group or role claim in the caller’s token. Your on-call team can clear queue statistics while everyone else can only read them.
  • Guardrails for production. Per-broker rate limiting, retries with backoff, concurrency caps, and an optional audit log of destructive operations. An assistant can’t flood a broker with requests.
  • Observable like any other service. Prometheus metrics, OpenTelemetry tracing, a correlation ID on every request, and an importable Grafana dashboard.

Works with the Tools You Already Use

Any MCP client can connect. Setup is documented for Claude Code and Claude Desktop. Deploy it as a standalone binary, a Docker image, or with the Kubernetes manifests that ship with it. Release archives come with checksums, and both the archives and the container image carry build provenance attestations.