AI agents & MCP
An agent connected to Snapi isn't just a script with API access — it's someone you can assign a card to. Give it an API key with a name, put that name in a card's "assigned to" field, and it can check the board for what's its and work it. Snapi's MCP server is live at https://api.snapi.ca/mcp; any MCP-compatible assistant — Claude, Cursor, Windsurf, n8n agents, and others — connects to it and can read and write your workspace through natural language: pages, records, kanban boards, comments, and file attachments.
What is MCP?
The Model Context Protocol is an open standard that lets AI assistants call tools on external systems. Instead of building prompt engineering hacks, you configure an MCP server once and the AI gets structured, reliable access to your data.
Snapi's MCP server exposes your pages, records, and boards as tools the AI can invoke. The assistant can answer questions like "How many leads came in this week?" or "Move record X to the Approved column" without you writing a single line of code.
Connecting
The server speaks MCP's Streamable HTTP transport at https://api.snapi.ca/mcp and authenticates with a Snapi API key sent as a header — either Authorization: Bearer snapi_... or X-API-Key: snapi_.... Get a key from Settings → API Keys and give it an access level: what the key may do is what the agent may do.
Claude Code (CLI)
claude mcp add --transport http snapi https://api.snapi.ca/mcp \
--header "Authorization: Bearer snapi_your_key_here"
Claude Desktop
Claude Desktop connects to remote servers through a local bridge. Add to claude_desktop_config.json:
{
"mcpServers": {
"snapi": {
"command": "npx",
"args": [
"-y", "mcp-remote", "https://api.snapi.ca/mcp",
"--header", "Authorization: Bearer snapi_your_key_here"
]
}
}
}
Other MCP hosts (Cursor, Windsurf, n8n, custom)
Any host that supports HTTP-based MCP servers just needs the URL and the auth header:
URL: https://api.snapi.ca/mcp
Header: Authorization: Bearer snapi_your_key_here
whoami tool to see the effective level. See API keys.Available MCP tools
Connected assistants get 71 tools. The agent discovers them automatically — this list is for planning which access level a task needs.
persist_records (default true — set false for a pass-through page that fires webhooks but saves nothing, see Webhooks); delete removes the page and its records. A table/chart field can set dataSourceEntityId to read a different page's records directly, no onload webhook required — see Field types. Adminassigned_to: "me" to list_records to see only what's assigned to you — how an agent checks its own queue instead of the whole board. Viewer+remove_card only unlinks it, never deletes). All content actions, all Editor+.remove_card so you can put it back with add_card_to_board. Entity-bound boards check just that entity; a freeform board checks the whole org instead, capped at 100. Editor+Example conversations
Once connected, you talk to your data naturally — and your agent can check its own work, not just yours:
You: What's assigned to you on the Support board right now?
Claude: Three cards. Two password-reset requests I can close myself, and one
refund request I've flagged for you — it's over your $200 auto-approve limit.
You: Handle the two, and move the refund to my column.
Claude: Done. Both reset tickets are closed and commented. The refund request
is now in "Needs Approval," assigned to you.
You: How many new leads came in this week?
Claude: You received 23 leads this week, up from 14 last week. Most came
through the website contact page.
You: Create a new contact: Bob Smith, bob@example.com, interested in enterprise plan.
Claude: Created. Record ID rec_8f2c9a for Bob Smith has been added to Contacts.
n8n AI agent integration
If you use n8n for automation, you can expose Snapi to n8n's built-in AI agent node via the REST API. Add Snapi as an HTTP Request tool in your n8n AI agent workflow:
- In n8n, create an AI Agent node.
- Add an HTTP Request Tool with the base URL
https://api.snapi.ca/api/v1. - Set
Authorization: Bearer snapi_...as a fixed header. - Describe the tool to the agent: "Use this to read and write Snapi records. GET /entities/:id/records to list, POST /entities/:id/records to create."
Security considerations
- Use a dedicated API key for each AI agent or integration — makes it easy to revoke access without affecting other systems.
- Give each key the lowest access level that does the job: a reporting bot gets Viewer, a triage agent Editor, a workspace-builder Admin.
- The MCP server inherits your organization's plan limits, just like the REST API. A runaway agent that creates thousands of records will hit your quota and stop (returning 402 errors).
- Record changes made by AI agents are written to each record's audit log, so you can review what the AI did.