Endpoint
Copy the server URL from Settings → API & MCP Access in the browser app.
Connect an MCP client
1
Open API and MCP Access.
Sign in at https://app.letsflow.ai. Go to Settings → API & MCP Access.
2
Copy the server URL.
Use
https://app.letsflow.ai/mcp (or copy from the Model Context Protocol (MCP) Server card).3
Add the server in your client.
Supported clients include Claude Code, Cursor, and Continue. Add a remote MCP server
with the URL above.
4
Authenticate with OAuth.
Complete the OAuth flow when the client prompts you. If you belong to more than one
organization, choose one during sign-in. Flow scopes the MCP session from that
selection in the Connect access token. Clients such as Cursor, Claude, and Codex do not
need a custom organization header.
5
Verify tools load.
The client should list Flow tools your principal is allowed to use. Tool visibility follows your
org role and capabilities.
How MCP fits with other surfaces
MCP exposes a small customer contract over Flow requests and approvals rather than mirroring internal API resources.
Authentication
Create service accounts and API keys on the same API & MCP Access page when you need machine access outside MCP.
What clients can do
Published MCP tools cover requester work and automation authoring or operation, subject to the signed-in user’s authorization. Requester tools:get_current_user: confirm identity, teams, and active request or approval keyssearch_requests: find requests or retrieve one by its public request keylist_approvals: list actionable approvals and open the exact approval in Flowlist_request_catalog: list every available request option and its input schemasubmit_request: submit a matching published request option or a general request
- List and retrieve authoring skills, approval procedures, integrations, action declarations, lint rules, connector health, automations, and runs.
- Lint complete automation TypeScript without saving it.
- Create a draft, save draft revisions and numbered versions, and run mocked validation review.
- Publish a saved version after Review and publication prerequisites pass.
- Start simulated or live runs asynchronously, inspect their state and safe logs, and request cancellation.
automation_url. You can use this link to view the automation in Flow. Your existing permissions determine what you can view and publish.
Save draft, Save version, Review, and Publish have separate outcomes:
- Save draft preserves source revisions while you edit.
- Save version records a numbered checkpoint, such as v1, using the same rules as Save in Flow.
- Review checks validation scenarios and required coverage for the exact saved revision.
- Publish makes a saved version available to your authorized audience after its prerequisites pass.
- Create a draft with
create_automation. - Save your draft source, input examples, and validation scenarios with
save_automation_revision. - Save a version with
save_automation_version, using the automation key and draft revision. - Run validation scenarios for the returned revision with
run_automation_validation_scenarios. - Confirm that scenarios and coverage pass with
list_automation_validation_scenarios. - Call
publish_automationwith that revision and your authorized audience.
status: "saved", revision, and version_number. Success confirms a saved checkpoint. Check Review results and publication state separately. Repeated saves of the same draft reuse its version. If updated connection contracts create a newer revision, validate the returned revision before publishing.
If publication returns setup_required with reason_code: "automation_version_not_saved" and next_step: "save_automation_version", call that tool and validate the returned revision. Retry publication with a new mutation key. Replaying a completed setup mutation returns its original receipt, including after saving a version.
You can complete this authoring path through MCP. Browser links provide optional navigation to the current automation. Opening a link does not save, publish, select a historic revision, or restore your requested audience.
MCP access mode
Organization administrators set MCP access mode on the Model Context Protocol (MCP) Server card in Settings → API & MCP Access. The mode controls what every MCP connection in the organization can do.
The mode is a ceiling, not a grant. It never lets a connection do more than that person can already do in Flow. Approval decisions stay in the authenticated browser. MCP cannot perform them.
A change applies to new MCP requests immediately. Flow records the change in the audit trail.
Troubleshoot connection issues
1
Confirm the server URL.
Confirm the server URL ends with
/mcp and uses https://app.letsflow.ai.2
Retry OAuth.
Retry OAuth if the client cached an expired token. Multi-organization users should complete
organization selection during that flow.
3
Check multi-organization linking.
If the client reports that your identity is linked in multiple organizations, disconnect and
reconnect so organization selection can be recorded on the new token.
4
Verify capabilities.
Verify your user has the capabilities required for the tools you expect (team admin or org
admin may need to grant access).
5
Check MCP access mode.
If write tools are missing for everyone, check the MCP access mode; an org
admin may have set it to read only.
6
Compare across surfaces.
Compare behavior with the same action in the browser or command-line interface (CLI)
to isolate client vs permission issues.
Legal pages for directory listings
When an assistant directory or marketplace asks for Flow legal URLs, use:
The privacy URL redirects to ServiceNow’s published privacy policy. Hosted MCP tools also advertise directory-safe titles and annotation hints (
readOnlyHint, destructiveHint, openWorldHint) on tools/list.