Skip to main content
Nora ships an MCP server so a coding agent (Claude Desktop, Cursor, Codex, and other MCP hosts) can manage your workspace: create Flows, add Blocks, run and inspect Traces, and publish — all from natural-language instructions. The same commands are available as the nora CLI. MCP is just the coding-agent-friendly way in.

​
What you can do from a coding agent

  • Create or edit a Flow: “Add a Slack Action to the support flow that pings #cx-alerts on refunds.”
  • Connect a data source: “Attach the /policies folder in Drive to the support agent.”
  • Debug: “Pull the last 5 failed traces from support-flow and summarize.”
  • Publish: “Snapshot the draft and publish with note ‘refund tool cleanup’.”
  • Explore: “List all Flows in this workspace and tell me which ones have unpublished changes.”

​
Setup — Claude Desktop

  1. In Nora, go to Settings → Security → API tokens and click New token. Copy it.
  2. Open Claude Desktop’s MCP settings file:
    • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
    • Windows: %APPDATA%\Claude\claude_desktop_config.json
  3. Add the Nora server:
  1. Restart Claude Desktop. In a new chat, you’ll see “nora” appear under available tools.

​
Setup — Cursor

Cursor supports MCP through the same config surface. In Cursor Settings → MCP, add:
Reload Cursor. Nora tools appear in your chat sidebar.

​
Setup — other MCP hosts

Any MCP-capable client works. Point it at @nora-ai/mcp-server via npx and pass NORA_TOKEN in the environment. Full command list matches the CLI reference.

​
What tools MCP exposes

The MCP server surfaces roughly the same operations as the CLI, plus a few MCP-native conveniences:
  • Flow management (list, get, apply, publish, rollback).
  • Block CRUD (agents, tools, actions, triggers, edges).
  • Data (documents, memory, retrieval, causal, pipelines).
  • Reliability (traces, feedback, approvals).
  • Workspace (providers, connectors, billing).
Each tool has typed inputs the coding agent auto-discovers.

​
Scoping tokens

Tokens issued from Personal API tokens are account-scoped — they can act on any workspace you belong to. For automation, prefer a workspace-scoped token: Settings → Security → API tokens → Workspace tokens. Workspace tokens can further be scope-restricted to specific operations (read-only, no-publish, etc.). Set at token creation.

​
What the coding agent can not do

MCP mirrors your permissions. If you’re a Viewer in a workspace, the agent can only read. Approval-gated actions still go through the approval flow — the coding agent cannot bypass them. Workspace policies (see Access) also apply. If publishing is restricted to CI, an interactive MCP session can’t publish either.

​
Rotating a token

Treat MCP tokens like passwords. Rotate them from the Nora settings page anytime; the old token stops working immediately.
To rotate:
  1. Issue a new token in Nora Settings.
  2. Update the NORA_TOKEN env in your MCP host’s config.
  3. Restart the MCP host.
  4. Revoke the old token.

​
Multiple workspaces

Configure one MCP server per workspace with distinct tokens:
The coding agent sees both, prefixes tool names by server, and can operate on either.

​
Troubleshooting

“nora not appearing in tool list” — check Claude Desktop / Cursor logs. Usually npx failing to fetch the package, or NORA_TOKEN unset. “401 unauthorized” — token revoked or expired. Re-issue. “Permission denied” — role or workspace policy issue. See Access.