Aplas DocsConnecting AI assistants (MCP)
Query and manage your software landscape using natural language through AI assistants
Aplas runs a Model Context Protocol server, so an AI assistant can read and change your estate on your behalf. Ask questions about your assets and how they connect, or have the assistant record what you have found — without writing queries or navigating the studio.
The assistant works in one workspace at a time, over the asset types you have defined. It sees the same data you do, limited to what your API key can reach.
What you can ask
Understanding what is there
- "List the asset types in my enterprise workspace"
- "How many applications are there, and how many are Retired?"
- "Search for anything matching 'payments'"
Following the connections
- "What does the CRM System connect to?"
- "Which relationship types does this workspace use, and how many of each?"
- "Show me everything related to the identity provider"
Recording what you find
- "Create an application called 'CRM System', vendor Acme, lifecycle Standard"
- "Set the CRM System's vendor to Contoso"
- "Group these twelve applications into a 'Customer' domain"
The assistant should call list_asset_types before writing, because your asset types decide which fields exist, which are required, and what a choice field will accept.
Available tools
Nine tools, and which ones work depends on your key's role:
| Tool | What it does | Key required |
|---|---|---|
| list_workspaces | The workspaces this key can reach. Called first — everything else needs a workspace. | ReadOnly |
| list_asset_types | The workspace's asset types, their fields, which are required, and valid choice options. | ReadOnly |
| count_assets | How many assets exist per type, optionally narrowed by a search term. | ReadOnly |
| count_relations | How many relationships exist per relationship type, including types with no rows yet. | ReadOnly |
| search_assets | Full-text search across the workspace's assets. | ReadOnly |
| get_asset | One asset by name — its fields and what it connects to. | ReadOnly |
| create_asset | Creates one asset. | ReadWrite |
| update_asset | Changes one asset's name, description or field values. Only the fields you send are altered. | ReadWrite |
| group_assets | Groups existing assets into named domains or areas. Strictly additive. | ReadWrite |
A ReadOnly key is refused the last three, and the refusal says which role was required.
Writes take effect immediately
There is no confirmation step over MCP. When an assistant calls one of the three write tools, the change is made and saved — unlike the assistant inside Aplas, which shows you a confirmation card first. Some clients ask you to approve a tool call before running it; that is your client's prompt, not ours, and it is the only chance to intervene. Use a ReadOnly key for any assistant you do not want changing data.
Every change is recorded in the workspace's change log, attributed to the API key that made it, so you can see afterwards what an assistant did.
MCP compared with the assistant inside Aplas
They share one set of tool definitions, so they understand your estate identically. They differ in what running a tool does, and in what is offered:
| In-app assistant | MCP | |
|---|---|---|
| Writes | Proposed — you confirm a card | Applied immediately |
| Workspace | The one you are in | Named per call, from list_workspaces |
| Importing a file | Yes | No — use the studio's import, or the API |
| Generating an estate or a view | Yes | No |
The three absences are deliberate: importing, generating and building views are interactive, and are better done where you can see the result before accepting it.
Guided workflows
The server offers a prompt, discoverable through MCP — in Claude Code it appears as a slash command, in other clients as a menu item:
| Prompt | What it covers |
|---|---|
| import-guide | How to get data into a workspace: what an asset type is, how relationships carry meaning, and when to use these tools versus the studio's bulk import. |
In Claude Code, invoke it with /mcp__aplas__import-guide.
Supported clients
Any client implementing MCP protocol 2025-11-25 over streamable HTTP works. These are the ones we test:
- Claude Code — Anthropic's CLI
- Claude Desktop, claude.ai and the mobile apps — as a custom connector
- Cursor — AI-powered code editor
- VS Code (GitHub Copilot) — via the MCP extension
Custom connectors reach us from Anthropic's cloud
When you add Aplas as a custom connector, Anthropic's infrastructure connects to our servers rather than your laptop doing it. That works because our MCP endpoints are on the public internet — no VPN or allowlisting needed at your end.
See the setup guide for the configuration for each client, and the API overview if you would rather call Aplas directly.