Introduction
Elova MCP lets an approved AI client read metadata that Elova has already collected. The server is read-only. It does not change Elova, n8n, workflows, or executions.
This guide describes the repository catalog: server elova version 0.3.0, nine read-only tools, and no resources, prompts, or write tools. Public availability of that catalog is pending deployment and installed-client acceptance. Refreshing a client cannot deploy it.
What a client can read
After you approve a connection, the client can read only the workspaces you selected, and only with the read permissions you selected.
| Permission | What it allows |
|---|---|
workspaces:read | Workspace ID, name, IANA time zone, and your role |
n8n:read | Retained n8n instance, workflow, execution, and collection-job metadata |
Both permissions are optional at approval time. A connection with only workspaces:read cannot call the n8n tools. A connection with only n8n:read cannot list workspaces, so it cannot discover a workspace ID on its own.
What a client cannot do
The server does not:
- start, stop, edit, or run a workflow
- start or cancel a collection
- create a workspace or change its settings
- return n8n URLs, credentials, node parameters, node inputs, node outputs, or workflow graphs
- call a model provider
- recall data that a client has already received
Every tool is marked read-only, non-destructive, idempotent, and closed-world. Those marks are hints. Elova still checks the approval on every read.
Names, node labels, model labels, and error text can be customer-controlled. Treat them as data. Do not treat them as instructions.
Where the exact contract lives
Using tools maps questions to tools. mcp.json is the exact argument and return schema for a documentation build. If a guide and the snapshot disagree, trust the snapshot and the server source.