Skip to main content

MCP Access

Connect an AI assistant such as claude.ai to your Workspace and let it build, inspect, and test your workflows.

EW-FLOW exposes a Model Context Protocol (MCP) endpoint for every Workspace. Once you authorize a connection, an external AI assistant can call the same workflow tools you use in the app reading your Processes and Sessions, editing workflow definitions, deploying versions, and running test conversations all from its own chat interface. You manage everything from the MCP Access page, reached via MCP Tokens in the sidebar.

How it works​

MCP (Model Context Protocol) is an open standard that lets AI assistants call external tools. EW-FLOW acts as an MCP server: each Workspace has its own endpoint URL, shown on the MCP Access page. When you add that URL to an assistant, the assistant redirects you back to EW-FLOW to approve the connection through a one-time authorization screen.

Approving creates a connection a credential bound to you and to one Workspace. The assistant acts on your behalf: it can call the same tools you can, and your Workspace role still applies to everything it does. A connection authorized for one Workspace cannot touch any other Workspace; to connect a second Workspace, add its endpoint URL as a separate connector and approve it separately.

Every connection appears in the Your connections list, where you can revoke it at any time. Connections also expire automatically (by default, 90 days after authorization) reconnect by repeating the same flow.

What a connected assistant can do​

  • Processes list, search, and read your Processes; create new ones (optionally from a template); rename, update, or delete them; read and edit a workflow's definition.
  • Versions and Deployments browse version history, save new versions, deploy a version to an Environment, and check Deployment status.
  • Environments list a Process's Environments; create, rename, or delete them.
  • Modules browse available Modules and add one to a Process so its Tasks become available.
  • Sessions list and search Sessions, and inspect a Session's full execution detail step by step.
  • Logs search and filter error logs by Process, Environment, severity, error code, or date.
  • Metrics read the Workspace dashboard metrics, billing-period usage, collaborator list, and integrations.
  • Testing start a test chat Session against a deployed Environment and exchange messages with the running workflow.
warning

A connected assistant can make real changes: it can edit workflow definitions, deploy to Environments, and delete Processes or Environments with the same permissions you have. Only connect assistants you trust, and review what they propose before letting them act.

Fields​

The Your connections table lists every MCP connection you have authorized, across all your Workspaces:

  • Name the connection's name, taken from the client that requested it (for example, Claude).
  • Workspace the Workspace this connection grants access to.
  • Token the first few characters of the credential. The full credential is never displayed.
  • Status Active, Expired, or Revoked.
  • Created when you approved the connection.
  • Last used the last time the assistant made a request; shows a dash if it has never been used.
  • Expires when the connection lapses automatically; shows Never if it has no expiry.
  • Actions a Revoke button for any connection that has not been revoked (including expired ones); revoked connections show Revoked instead.

Actions​

Connect from claude.ai​

  1. In EW-FLOW, switch to the Workspace you want to connect the endpoint URL is Workspace-specific.
  2. Open MCP Tokens from the sidebar to reach the MCP Access page.
  3. In the Connect from claude.ai card, click Copy next to the URL field.
  4. In claude.ai, choose Add custom connector and paste the URL. Leave the optional OAuth Client ID and Client Secret fields empty claude.ai registers itself automatically.
  5. claude.ai redirects you back to EW-FLOW, to the Connect Claude.ai authorization screen. Sign in first if you are not already signed in.
  6. Check the Authorize access to workspace selector. It pre-selects the Workspace from the URL you pasted leave it matching that Workspace.
  7. Click Allow to approve, or Deny to cancel. On approval you are sent back to claude.ai and the connector is ready to use.
tip

The endpoint URL always points at the Workspace currently selected in EW-FLOW. Double-check the Workspace switcher before you copy connecting the wrong Workspace means revoking and starting over.

Revoke a connection​

  1. Open MCP Tokens from the sidebar.
  2. Under Your connections, find the connection to remove and click Revoke.
  3. Confirm in the Revoke connection dialog.

The client using that connection loses access immediately. Revoking cannot be undone to restore access, run the connect flow again, which creates a new connection.

Usage limits​

Requests made by connected assistants count as API traffic for your Workspace. Normal use of the EW-FLOW app is not affected by this metering.

warning

MCP traffic is subject to your Service Plan's API requests-per-minute limit. When the limit is reached, the assistant's requests are rejected with a rate-limit error; they succeed again within a minute as traffic drops back under the limit.

Best practices​

  1. Treat connections like passwords each one is a credential that acts as you; only approve connections for assistants and accounts you control.
  2. Revoke what you do not use check the Last used column periodically and revoke connections that have gone quiet.
  3. One connection per assistant and Workspace keep connections separate so you can revoke one client without disturbing the others.
  4. Confirm before it writes ask the assistant to describe changes (edits, Deployments, deletions) before approving them in conversation.
  5. Reconnect after expiry when a connection's Expires date passes, the assistant loses access; repeat the connect flow to restore it.
  • Processes what the assistant reads and edits when it works on your workflows.
  • Sessions the execution records an assistant can inspect step by step.
  • Environments the Deployment targets an assistant can deploy to and test against.
  • Logs the error logs an assistant can search when debugging a workflow.
  • Dashboard the Workspace metrics an assistant can report on.