MCP onboarding
Connect. Ask to onboard. Check.
Your assistant prepares an OpenTelemetry plan, instruments your MCP server with your approval, and checks the result. No proprietary SDK. No extra tracing platform.
01 / Connect Flowlines
Add Flowlines to your assistant.
Use the Use with button to connect the Flowlines MCP, or install the official Flowlines plugin. Sign in when prompted, then open your MCP server repository in your coding assistant.
New to Flowlines? Create your account and workspace. The get-started page also creates your namespace API key when you need one.
Claude Code
claude plugin marketplace add flowlines-ai/plugins && claude plugin install flowlines@flowlinesRun /mcp inside Claude Code, select flowlines, and complete the OAuth sign-in.
Codex CLI
codex plugin marketplace add flowlines-ai/plugins && codex plugin add flowlines@flowlinesThen run codex mcp login flowlines to complete the OAuth sign-in. Start a new session or task if the plugin is not yet available.
The plugin includes the hosted MCP connection at https://api.flowlines.ai/mcp. If you already configured an MCP server named flowlines manually, keep one connection to avoid duplicates.
Copy the install commands from manual setup › or inspect the official plugin source. These instructions match desktop plugin v0.5.0.
02 / Ask to onboard
Tell your assistant what to do.
Onboard my MCP server to Flowlines.
Your assistant calls the onboarding MCP tool to get an instrumentation plan for your namespace, then checks your repository. If setup is needed, it guides you through choosing your workspace and what to record.
Onboarding is an MCP tool workflow. It does not require a separate instrumentation skill.
03 / Approve the plan
Review it. Let your assistant implement it.
The plan uses OpenTelemetry and fits your server. Review the proposed code changes and data capture, then give your assistant the go-ahead to instrument the MCP and run the tests.
It guides the required configuration and tells you when to start the server. You do not need to wire up a proprietary SDK or another tracing service yourself.
04 / Check onboarding
Ask Flowlines to verify the setup.
Check my Flowlines onboarding and fix anything missing.
Your assistant runs the test calls, calls check_onboarding to check the received telemetry, and fixes any reported gaps. Once the checks pass, follow its deployment instructions and check again after deployment.
onboarding and check_onboarding return a plan and verification results; your coding assistant applies the approved repository changes. A successful sign-in alone does not verify instrumentation.
That is the onboarding flow. Connect, approve the OpenTelemetry plan, and verify.
After onboarding
Start learning from your MCP.
See users, requests and tool journeys. Open a finding to inspect the sessions behind it.
Explore MCP analytics ›Desktop plugin v0.5.0
Choose the skill for your task.
The GitHub marketplace plugin includes these five skills. Invoke them in a new session after installation and sign-in.
| Task | Claude Code | Codex |
|---|---|---|
| Review usage, signals and outcomes since the last review. | /flowlines:flowlines-weekly-review | $flowlines-weekly-review |
| Compare outcomes before and after a release. | /flowlines:flowlines-release-check | $flowlines-release-check |
| Investigate a signal, complaint or failing session. | /flowlines:flowlines-investigate-session | $flowlines-investigate-session |
| Define a user cohort and compare it with a baseline. | /flowlines:flowlines-cohort-builder | $flowlines-cohort-builder |
| Diagnose missing telemetry or MCP connection problems. | /flowlines:flowlines-doctor | $flowlines-doctor |
For Cursor or another MCP client, use the hosted connection and ask to onboard in natural language. The commands above are specific to Claude Code and Codex.
The public ChatGPT and Codex submission bundle built from v0.5.0 contains the four analysis skills and excludes flowlines-doctor. A published directory installation can differ from this source bundle; check the skills available in your client. See the public plugin guide.
For agent clients
Connect to the live MCP.
The Streamable HTTP endpoint is https://api.flowlines.ai/mcp. Use an MCP client and sign in to Flowlines with OAuth. A browser GET is not an MCP handshake.
After sign-in, initialize the connection and use tools/list for the current tool schemas. Then ask to onboard. The Free plan includes 100 analyzed MCP sessions per month, with no credit card or sales call.
OAuth currently advertises the email identity scope, not separate read-only or write scopes for tools. Workspace and namespace access still apply. Review requested actions before approving changes.
If instrumentation needs a namespace API key, your assistant guides you to create one in Flowlines. Keep it in your local environment or deployment secret manager. No shared public sandbox is advertised; use synthetic data in a workspace you control.
- MCP server description in the published MCP Registry schema format. This describes our endpoint; it does not claim a registry listing.
- OpenAPI description of the MCP transport. Native MCP tool discovery remains authoritative; this is not the internal REST API contract.
- Protected-resource metadata and authorization-server metadata.
Common setup questions
Do I need a proprietary Flowlines SDK?+
No. Your server uses OpenTelemetry instrumentation. The Flowlines MCP or plugin guides the setup; it is not a proprietary runtime SDK you need to add to your server.
Do I need another tracing platform?+
No. You can send OpenTelemetry telemetry directly to Flowlines. There is no requirement to sign up for another observability service or migrate your existing tracing stack.
Do I need both the MCP connection and the plugin?+
No. Connect the Flowlines MCP to your assistant, or use the official plugin that includes the connection. To apply the instrumentation plan, use a coding assistant with access to your MCP server repository.
Does connecting Flowlines automatically instrument my server?+
No. Ask to onboard, then approve the proposed code changes. Your assistant implements the plan in the repository you choose.
What if I do not have a workspace or namespace API key?+
Open app.flowlines.ai/get-started to create your account, workspace and namespace API key. Keep the key in your local environment or deployment secret manager, not in chat.
What does the onboarding check do?+
Your assistant runs the plan's test calls and asks Flowlines to check the received telemetry. It fixes reported gaps and checks again. A connection alone is not a passed instrumentation check.