MCP
The MCP screen lists every tool that Axelix Master exposes through its bundled Model Context Protocol server. Each card is one tool an AI agent connected over MCP can invoke against any registered Spring Boot instance.
When the MCP tab appears
The MCP link in the navigation bar is rendered only when the MCP server is enabled in Master. The gate is the
axelix.master.mcp-server.enabled property, which defaults to true. In deployments that disable the server, the link
is hidden and /mcp-server is unreachable. See MCP server
configuration for the property.
The MCP server itself authenticates AI agents separately from the browser session — credentials, the supported Basic
and Bearer schemes, and per-tool authority gating are covered in MCP server
authentication. This page
describes only the read-only catalog UI, it does not change which tools an agent can call.
The tool catalog
The catalog is a grid of cards. The search field at the top filters cards by title, and the counter on the right shows
<matches> / <total>. The search matches anywhere in the tool title and is case-insensitive.
Every card has the same three parts:
- Header with the tool title and a coloured status dot — green when the tool is registered with the MCP server, red when it is currently unavailable.
- Description — the short blurb the tool advertises to MCP clients. Long descriptions are clipped to a fixed height; hover the description to see the full text in a tooltip.
- Footer with annotation tags — see Tool annotations below.
Tool annotations
Each tool declares four boolean hints that tell MCP clients what kind of side effects they should expect when calling it. The card footer surfaces two of them inline and exposes all four through a popover.
Inline tags (always visible):
- Read-only — the tool does not change any state in the target Spring Boot instance.
- Idempotent — invoking the tool more than once with the same arguments has the same effect as invoking it once.
Each tag is coloured green when the hint is true and orange when it is false. The visible value is the tool's own claim — not a guarantee about what the
underlying endpoint actually does.
Full annotations popover — click the burger button on the right of the footer to open a panel showing all four hints with their boolean values:
- Read-only — the tool does not change any state in the target Spring Boot instance.
- Idempotent — invoking the tool more than once with the same arguments has the same effect as invoking it once.
- Destructive — the tool may delete data or otherwise apply changes that cannot be undone from the UI.
- Open-world — the tool reaches outside Master and the registered Spring Boot instances (for example, an external HTTP call).
What is missing from this screen
The MCP page is a discovery surface, not a runner. From here you cannot:
- Invoke a tool against a Spring Boot instance — that happens from your MCP client (Claude Desktop, Cursor, a custom agent, etc.).
- Enable or disable individual tools — the catalog reflects what the master registered at startup.
- See past invocations or per-tool metrics.
To use the catalog, point your MCP client at <master-base-url>/api/mcp and authenticate as described in MCP server
authentication.
An Access Log for MCP tool invocations is on the roadmap and will ship with Axelix Enterprise. Both the Access Log and Axelix Enterprise itself are still in development, so for now the MCP screen remains a read-only catalog without history or per-tool metrics.