Triage Broken Sessions From Your Phone With a Manus Cue Connector for Session Analysis

Triage is where a Manus Cue connector for session analysis earns its place: the agent reads the issues Flowsery found in your session recordings, ranks them, and hands you the three recordings worth watching. Cue is the personal-agent app Manus launched on 28 September 2026, built on the same infrastructure as Manus. Every call triage needs is a read, and exactly one session call writes.

The short version: build the connector from the Manus Cue connector for web analytics guide with a Contributor token, ask for open critical issues on one website, and keep update_issue_status for yourself. Cue documents no approval prompt, so the token role is what stops the write.

What does a Cue agent see when it opens a Flowsery issue?

A Cue agent sees the issue record Flowsery built from your recordings, not the recording itself. Each issue carries a title, description, severity, status, sessionsCount, firstSeenAt, lastSeenAt, ordered stepsToReplicate and a sampleRecordingId. Severity takes low, medium, high or critical.

get_issue (GET /issues/{issueId}) adds three things the list leaves out. The occurrences array gives each flagged moment a description, a severity and an atSeconds offset into the recording. The sessions array lists every recordingId behind the issue. The comments array shows what your team already said, which separates a rage click cluster nobody has seen from one a developer looked at on Thursday.

The issue also carries externalTicketProvider, externalTicketKey and externalTicketUrl when someone has linked a ticket. Tell the agent to read those before it suggests filing anything. A Cue agent has its own email address, and a duplicate bug report sent from it is still a duplicate.

Can a Cue agent watch the session replay?

A Cue agent cannot watch a Flowsery replay through the connector, because the API returns JSON and a replay is a recording of the page. The agent reads the summary: which moment broke, at which second, in how many sessions. It cannot tell you the submit button looked greyed out.

Cue adds a temptation Muse and Grok do not. Each Cue agent has "its own email, phone number, wallet, and computer", so the obvious workaround is to log the agent into the Flowsery dashboard on its computer and let it watch. Skip it. A dashboard login carries your role, and if you are an Admin, the agent gets every write you have, member management included. The connector with a Contributor token gives it the issue data and nothing else.

Let the agent pick the recordings. Then watch them yourself, starting at the atSeconds it names.

What do you paste into Cue for issue triage?

The triage prompt names the website, the filter, the fields to print and the write ban in one message. The ban belongs in the first message, because Manus has not documented a Cue approval screen that catches it later.

Use the Flowsery connector. Call list_websites and pick flowsery.com.
 
Call list_issues with status=open and severity=critical, sorted by severity.
For the top three, call get_issue and give me:
title, sessionsCount, lastSeenAt, stepsToReplicate,
the recordingId with the earliest atSeconds, and any externalTicketUrl.
 
Print severity and sessionsCount next to every summary.
Do not call update_issue_status. Do not email this to anyone.
If an issue has no sampleRecordingId, say so. Do not pick another recording.

"Print severity and sessionsCount next to every summary" is the line that keeps the ranking honest. list_issues sorts by severity unless you pass sort=recency, and a written summary reorders the list by whatever reads best.

Which session calls read, and which one writes?

Four Flowsery calls matter for session analysis, and only PATCH /issues/{issueId} changes data. The MCP server exposes it as update_issue_status.

A token created under Contributor turns that write into 403 permission_denied, because Flowsery's flowsery.write permission belongs to Admin and Editor only. Flowsery's Create Token form marks Contributor "Recommended for AI agents" and describes it as "Same access as Viewer here", which is "Sees everything, changes nothing".

get_visitor is a read with a privacy cost. It returns a profile with userId, name and email for visitors your site has identified. That data belongs in the chat with you. An agent with its own outbox does not need it for triage, so leave the call out unless you are chasing one specific user's bug.

Who changes an issue's status?

A person changes an issue's status, after watching the recording. PATCH /issues/{issueId} moves an issue between open, in_progress, resolved and suspended, and Flowsery's spec says suspended issues "disappear from the default list". An agent tidying the backlog on its own judgment produces a shorter list with the same broken checkout underneath it.

The Meta Muse connector for session analysis leaned on Meta's Allow once prompt for this. Runtimewire reported on 28 September 2026 that the Cue announcement "does not explain in detail how users approve individual transactions". With no documented gate on the Cue side, the gate moves to Flowsery: a Contributor token for the agent, and an Editor or Admin login for you.

Manus's group chats fit this split. Agents in a Cue group chat share a goal and "hand work to each other", while "you set the direction and make the final call". Let a triage agent post the ranked shortlist into the group. The status change stays with you, in the Flowsery dashboard, after the replay.

Can a Flowsery alert start the triage?

A Manus Triggered task starts work when something happens in a connected service, and Manus names "a Slack message" among its examples. Flowsery posts issue alerts to the Slack channel you pick, so a Triggered task watching the alert channel can pick up the new issue and call get_issue on it. Manus also lists GitHub as a trigger source, and Flowsery opens its fixes as draft pull requests in GitHub.

Two limits apply as of 29 September 2026. Triggered tasks are a Manus feature, and Cue documents no triggers of its own. Manus has not documented whether an automation reaches a custom MCP server either. Build the trigger in Manus, point it at one test issue, and confirm the get_issue call returns data before you connect a live alert channel.

For the same loop with the instructions already written, the Grok bot template for session analysis runs the identical call chain from a chat.

What breaks in the first week?

Five failures show up in the first week of session triage, and the fourth one looks like an answer.

A 400 on the first call. The agent skipped the website selector. A workspace token spans every website in the workspace, so every call needs websiteId or domain. Tell it to call list_websites first.

An invented visitor ID. GET /visitors/{visitorId} takes the visitor ID from breakdowns and the tracking script, and no issue payload carries one. A guessed ID returns 404, the same response a real visitor with no data gets.

An issue with no recording. sampleRecordingId is nullable. The prompt above tells the agent to say so instead of substituting a nearby recording.

A reordered ranking. The summary reads well and the severity order is gone. Keep severity and sessionsCount printed beside every line.

The free-trial wall. On a Flowsery free trial, issues beyond the first 10 answer "Upgrade to view this issue". The agent reports ten issues and stops, which is the correct behavior and looks like a bug.

The session replay tool is where you watch the recordings the agent picked.

Frequently Asked Questions

Can a Cue agent close a Flowsery issue on its own?

A Cue agent holding a Contributor token cannot close an issue. update_issue_status needs the flowsery.write permission, which only Admin and Editor hold, so a Contributor token gets 403 permission_denied. Manus has not documented a Cue approval prompt as of 29 September 2026, so the token role is the control that holds.

Does Cue support MCP for session analysis?

Manus supports custom MCP servers under Settings, Integrations, Custom MCP Servers, Add Server, and Cue runs on the same infrastructure. Manus has not documented a Cue screen for MCP. The web analytics guide covers the form and the in-chat fallback.

What does a Cue agent see about a visitor?

get_visitor returns an identity block with location, browser, OS and device, plus activity, revenue and a profile carrying userId, name and email once your site identifies the visitor. Flowsery stores no IP addresses and sets no cross-site identifiers. Leave the call out of routine triage.

Does this replace opening the replay?

The connector ranks and narrows, and the recording shows the reason. Let the Cue agent pick three sessions and the second each one breaks at, then watch those three. Aggregate traffic questions belong in the analytics reporting guide.

Why is a suspended issue missing from the agent's list?

GET /issues leaves suspended issues out of the default response. Ask for status=suspended to see them. If one vanished and nobody on your team moved it, check which tokens hold Editor or Admin at flowsery.com/api-tokens.

Which Flowsery plan do I need for Cue triage?

Both Flowsery plans include the API, the MCP server and AI session analysis. Team costs $250 a month with 3-month replay retention, and Pro costs $500 a month with 12-month replay retention. Both start with a 14-day free trial that needs no card.

What is Manus Cue?

Cue is the personal-agent app Manus launched on 28 September 2026, built on the same infrastructure as Manus. Each Cue agent has "its own email, phone number, wallet, and computer". That is why the token role matters: the agent has an outbox of its own.

How does list_issues sort its results?

list_issues sorts by severity unless you pass sort=recency. Severity takes low, medium, high or critical. Ask the agent to print severity and sessionsCount next to every summary so a written recap does not reorder the list.

Why does get_visitor return a 404?

The agent most likely guessed the visitor ID. The ID comes from breakdowns and the tracking script, and no issue payload carries one. A guessed ID returns 404, the same response a real visitor with no data gets.

How many issues does a Cue agent see on a Flowsery free trial?

The first 10. Issues beyond that answer "Upgrade to view this issue", so the agent reports ten and stops. That is correct behavior, even though it looks like a bug.

Create a Contributor token at flowsery.com/api-tokens, run the triage prompt in Cue, and open the top recording yourself.

Sources: Introducing Manus 2.0 (28 September 2026), How can I use Manus Connectors?, Manus custom MCP docs, Manus Automations docs, Runtimewire (28 September 2026), the Flowsery OpenAPI spec and the Flowsery API reference. Checked 29 September 2026.