The tool calls that actually ran.
MCP: observe() records host tools/call. Personal agents: wrap records calendar, mail, and booking jobs.
What ran
Livefetch
A host called the MCP. The call failed.
get_calendar
A person asked. The wrap found the meeting.
book_table
A person asked. The wrap checked. It was not there.
Tool calls and jobs from real use.
What ran. How long. Whether it found it.
observe() on an MCP. wrap on a personal agent. A second look when you name a read.
MCP
observe() records tools/call on your MCP.
What we watch
When Cursor or another host calls a tool on your server.
How it gets in
pip install agent-status-sdk. Call observe() after tools are registered. Do not use track().
Write, then read
lookup= is the write tool then the read tool. We pass the new id on the same request.
Personal agents
wrap records the jobs the agent ran for a person.
What we watch
Calendar, mail, booking. The job, not the agent.
How it gets in
Agent() and wrap_tools on the dispatcher.
The second look
After a job the wrap finds a list or get sibling on the machine. You do not set a second URL.
Found or miss
The second look says whether the work is there.
Found
The read came back with the thing the write just made.
Miss
The write ran. The read did not find it.
No map, no second call
If there is no read pair, we keep the job and the time. We do not invent a check.
On the page
Each row is a job. Threads and alerts go on top of that.
The row
Job name, how long it took, and whether the check found it.
The thread
Jobs from one session stay together.
A miss after a ship
A miss on a new release is the cut. Email or webhook, same channels you already use.
What we keep
The wrap checks on your machine. We never sign into Google.
Three things
The job name, how long it took, and whether it found it.
What we do not keep
Names, the chat, or calendar contents.
The tools still run
If the wrap misses the call, the tool still runs. We do not block the job.
FAQ
Common questions
| Question | Answer |
|---|---|
| Are MCP and personal agents the same thing? | No. MCP uses observe() on the server. Personal agents use Agent() on the dispatcher. Same dashboard tab. Different function. The wrong one is rejected. |
| Do I need an SDK? | For MCP live traffic, yes. observe() on the server. For a personal agent, wrap the dispatcher. |
| What is lookup? | The write tool then the read tool. On MCP you name the pair. On a personal agent the wrap finds a list or get sibling on the machine. |
| Do you log into Google? | No. The wrap checks calendar and mail on your machine. We never sign into Google. We do not keep names, the chat, or calendar contents. |
