11. Turning It into an AI Assistant (MCP)¶

Connect an AI assistant like Claude or ChatGPT to eeumsae and you can build automations just by describing them.

"Every morning at 9, pull yesterday's orders from the orders API, summarize them, and send it to the Slack #ops channel"

The AI hears that, writes the workflow YAML, validates it, saves it, runs it and checks the result. The connection uses a standard called MCP (Model Context Protocol).

Step 1 — Issue an API key¶

Go to API Keys in the sidebar.
Create a new key.
Copy the key — it is shown exactly once — and keep it somewhere safe. Once you close the dialog you cannot see it again.

This key can reach the entire workspace. If it leaks, delete it on the API Keys screen immediately and issue a new one.

Step 2 — Connect it to your AI assistant¶

The connection instructions are on the same API Keys screen. The server address is:

https://api.eeumsae.com/mcp

Authentication goes in the X-API-Key header.

Claude Code (CLI)¶

Run this once in a terminal and it is registered.

claude mcp add --transport http weaver https://api.eeumsae.com/mcp \
  --header "X-API-Key: <your-key>"

Per-project configuration (`.mcp.json`)¶

Add it to .mcp.json in your project root.

{
  "mcpServers": {
    "weaver": {
      "type": "http",
      "url": "https://api.eeumsae.com/mcp",
      "headers": {
        "X-API-Key": "<your-key>"
      }
    }
  }
}

If you commit this file to git, do not put the key in it. Move it into an environment variable or a personal config.

Claude Desktop¶

Claude Desktop does not support HTTP header authentication directly, so it uses the mcp-remote bridge. Add this to claude_desktop_config.json.

{
  "mcpServers": {
    "weaver": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://api.eeumsae.com/mcp",
        "--header",
        "X-API-Key:<your-key>"
      ]
    }
  }
}

Other MCP-capable tools¶

The API Keys screen has one-click add buttons for the major tools. For a tool without a button, register the server URL and the X-API-Key header manually.

Step 3 — Check the connection¶

Ask your AI assistant:

"Check my eeumsae workspace"

If it comes back with the workspace name, you are set.

What the AI can do¶

Once connected, the AI can use the tools below.

Looking around

Tool	What it does
`get_workspace`	Check the workspace
`list_integrations` / `get_integration`	List available integrations and their input fields
`list_connections` / `list_connection_resources`	List connected services and their channels

Building

Tool	What it does
`validate_workflow`	Validate YAML without saving
`create_workflow`	Create a new workflow
`update_workflow`	Edit an existing workflow (full replacement)
`list_workflows` / `get_workflow`	List and inspect

Running

Tool	What it does
`execute_workflow`	Run it right now
`get_execution`	Inspect the result node by node
`list_executions`	Execution history
`cancel_execution`	Cancel an execution

The AI usually works in this order: check integrations → write YAML → validate → save → run → check the result → (if it failed) fix and repeat.

⚠️ Sending and payments ask for approval once¶

If the workflow contains a node that actually affects the outside world — like posting to Slack — execute_workflow does not run it straight away on the first call.

Instead it returns a list of "here is what will happen" and asks the user to confirm.

This workflow will do the following:
  · notify (slack_post_message) → send a message to channel C0123ABCDEF

Proceed?

It only runs once a human approves. That safeguard exists so an AI cannot accidentally blast messages at your customers. Read the list, and if something looks wrong, do not approve it.

Getting good results¶

✅ Be specific¶

"Every day at 9 AM Korea time, pull yesterday's orders from https://api.myshop.com/orders, compute total revenue and order count, summarize it in 3 lines with GPT, and send it to the Slack #ops channel. The API token is in the secret SHOP_API_TOKEN."

Useful requests¶

What you want	Just say
See the list	"Show me my workflows"
Find why it failed	"Look at why the last `revenue-report` run failed"
Edit	"Add 3 retries to that workflow"
Test	"Run it once now and tell me the result"
Check integrations	"What integrations can I use right now?"

Worth knowing¶

Validation is free. validate_workflow does not save, so it is fine to tell the AI "just validate it".
Validation does not catch expressions. A wrong ${...} path only shows up when it actually runs. Run it at least once.
Edits are rejected while an execution is running. Wait for the running execution to finish, or cancel it, and try again.
create_workflow fails on a duplicate id. Use a different id, or fix it with update_workflow.

Using the API directly¶

You can also call the REST API directly, without MCP. Put the same API key in the X-API-Key header.

curl -H "X-API-Key: <your-key>" https://api.eeumsae.com/workflows

The JSON Schema for workflow YAML is available below. Wire it into your editor (yaml-language-server) and you get autocomplete and validation while you write.

https://api.eeumsae.com/schema/workflow.json

Next → 12. Recipes