Observed
The operations and results present in the telemetry.
MCP session analytics groups related tool calls into one attempted job. Flowlines brings together the sequence, the use case and how the job ended, so you can see more than the status of each request.
100 analyzed MCP sessions per month, free. No credit card. Connect your assistant, then instrument your server.
Compare a completed request, a partial result and a session with no known outcome.
ChatGPT3 tool calls1.6 sGoal
Send last month’s revenue report to the board
Put together last month’s revenue by plan and send it to the board as a report.
Thinking
I’ll list the invoices, aggregate revenue by plan, then export the result.
A CSV was exported. The board report the user asked for was not produced.
Can you send that as a PDF instead? The board won’t open a CSV.
Product need detected · PDF export · 31 sessionsExecution map
An MCP session is the group of tool calls that belong to one attempted job, such as preparing a customer brief, from the first call to how the job ended. Flowlines groups calls into sessions this way and labels any session it infers.
For product analytics, it is not automatically the same as a network connection, transport session or single trace. The MCP 2026-07-28 specification removed protocol-level sessions from the modern stateless core, and older integrations may still use a transport session identifier. Neither case automatically groups a user's work.
An inferred session is an approximation: related work can be split, or separate requests can be grouped together. Calls without enough context may remain unassociated.
MCP's official July 2026 specification release ↗Glossary: MCP session ↗
Yes, at the level your server observes: Flowlines shows the ordered tool calls in a session, with timing, errors, captured results and how the job ended, but not the assistant's private conversation.
Work in another server, uncaptured calls and the assistant's final answer can remain outside that view. If you need to verify a complete task, first check which parts your instrumentation actually covers.
| View | What it contains | What it does not guarantee |
|---|---|---|
| Tool-call log | A record of an individual operation and its available result. | The larger job or what happened next. |
| Analytics session | Related observed calls, what the person was trying to do and how it ended. | Every step, every server or a verified final result. |
| Conversation transcript | Messages captured by the assistant application when supported and permitted. | That a server-side integration receives those messages. |
Flowlines groups sessions into use cases, such as preparing a customer brief or updating a record, and shows how often each was attempted and how it ended: accomplished, partial, failed or unknown.
A tool name is not a goal. search_docs can support onboarding, debugging or a security review. Use cases show the jobs people bring to your server, which ones it handles well and which keep ending in a partial result or a workaround. Validate the examples before turning a pattern into a roadmap decision or commercial signal.
| Use case (synthetic) | Sessions | What to check |
|---|---|---|
| Prepare a customer brief | 38 | Partial results caused by stale activity data. |
| Investigate an account | 26 | Repeated searches for similarly named companies. |
| Update a customer record | 20 | Whether the note reached the right account. |
Flowlines separates call success, whether an operation completed technically, from how the session ended, whether the requested job was accomplished; the two should never be silently substituted for one another.
Consider a customer brief: search for an account, retrieve its details, then fetch recent activity. Every call can succeed while the activity is too old to answer the user's question. A partial result makes that gap visible. Without a known outcome or other evidence, the final result remains unknown.
Inspect repeated searches, alternative paths and errors in context. Repetition may be a recovery attempt, but it can also be normal pagination or a legitimate second lookup. A sequence reveals where to investigate; it does not prove the cause by itself.
A Flowlines session connects the available evidence; its completeness depends on the instrumentation, identity and captured context, not simply on the number of tool calls.
| Context | What it adds | Boundary |
|---|---|---|
| User and client | Who or what generated the activity. | Only when known; shared identities can represent several people. |
| Use case | What the person was trying to do. | Not the entire conversation or the model's hidden reasoning. |
| Tool sequence | Operations, timing, repeated calls and errors. | Only the calls observed by the integration. |
| Arguments and results | The captured request and client-visible response. | Availability depends on capture; payloads can contain sensitive information. |
| Outcome | How the job ended. | A claim to review against the calls, not independent verification. |
| Session grouping | A path through related calls. | Inferred grouping is labeled and may be incomplete. |
A session may end accomplished, partially completed or failed. Read that outcome alongside the observed calls. An outcome that conflicts with available evidence deserves investigation, but a missing outcome is not proof of failure.
Show outcome coverage with any completion metric. Ten successes out of ten sessions with a known outcome tells you little about another ninety sessions without one. Do not classify inactivity as abandonment unless your product has evidence supporting that interpretation.
The operations and results present in the telemetry.
How the session ended, when that is known.
The context or final result the integration did not capture.
The Flowlines plugin gives your assistant the setup workflow. Connecting the assistant does not automatically instrument your MCP server.
MCP OpenTelemetry instrumentation guide ↗MCP observability product overview ↗The MCP analytics measurement guide ↗
An MCP session is the group of tool calls that belong to one attempted job, from the first call to how the job ended. It is not automatically the same as a transport connection or a single trace.
Yes, at the level your server observes. Flowlines shows the ordered tool calls, timing, errors, captured results and how the job ended. It does not include the assistant's private conversation, other servers or the final answer.
Flowlines groups sessions into use cases and shows how often each was attempted and how it ended: accomplished, partial, failed or unknown. Inspect example sessions before acting on a pattern.
When suitable context exists, Flowlines can infer a session. Inferred grouping is approximate and labeled. Calls without enough context are not silently presented as complete journeys.
No. The result is unknown. Missing telemetry, a client that did not send a result or work completed outside the server can all explain the absence.
Start with the traffic you already have
Connect Flowlines and take the next step with your assistant.