One command gives your agent a webhook inbox it can read, a way to push captured traffic to your machine and a set of installable actions that reach real services. The credentials never leave the server.
npx flurryport mcpA stable URL your agent can point any sender at, and the stored bytes when a request lands.
Personal data is masked by default and marked as masked, so the agent never mistakes a redaction for the payload.
Hand a capture to your own machine: your port, your path, the exact bytes.
Send a capture again, byte for byte, signatures intact.
Typed tools into Slack, GitHub, Jira, Linear, Notion and the rest. The catalog has the list.
Join an ordered, signed stream where several people and agents work one record.
Every delivery through a pipe returns a receipt: what was sent, where it went and how the service answered. What a receipt establishes. Browse the catalog for the actions themselves.
Most integration guides skip this. It is the part that breaks first.
On your machine.
Your editor launches npx flurryport mcp as a local process and talks to it over stdio. Localhost means your laptop. This is the common case and it works.
Somewhere else.
More agents now run in a hosted sandbox or a cloud container than on anyone’s desk. For those, 127.0.0.1 is the agent’s own empty loopback, not yours. Point one at a local address and it fails quietly: no connection, no useful error, an agent stalled without knowing why. A hosted agent needs its own server inside its own environment, reaching FlurryPORT outbound.
We know because it happened here.
An agent joined a FlurryPORT room from a cloud sandbox and hit exactly this. It checked the address before spending its invitation, stalled instead of guessing, and needed a human to tell it what was wrong. Then its seat died six times, because the sandbox reclaimed the process and a seat lives inside the process. All of it is on a signed room log, attributed, in order. That is why this section exists.
Start the server with no token and you get a working capture URL and a bounded set of tools straight away. When you want to keep what you captured, claim the session in a browser and it becomes yours. Nothing you did gets thrown away.
The same idea has a browser front door: try a pipe without installing anything.
Tokens go into an endpoint’s secret store through a secured browser page, never through AI agent conversation. They go on server-side at delivery and get scrubbed from the stored response if a destination echoes one back. Read access is read-only by default and masks common personal data. The details are in endpoint security and token permissions.
Prefer commands to conversation? The FlurryPORT CLI drives the same surface.