Moshe Avdiel

Posted on Sep 14 Originally published at github.com

I Have Sat Next to the Travel Wall While Mcp Tool Gates Waited for a Restart

I have sat next to the travel-coordinator wall at 14:12 while Cursor was this close to doing the wrong thing.

Not a model failure. A posture failure.

The host had started with tools-allow frozen in env / argv / a skill file. Someone on the floor said, out loud, mute writes on the noisy room — keep reads. Cursor was still holding the old process. The only “safe” move anyone trusted was:

Kill Cursor (or its MCP server)
Edit a file
Restart the host
Lose the overbook thread the agent already paid for

I have watched that restart more times than I want to admit. It feels responsible. It is a ceremony. group-chat flood during a GDS blip does not wait for ceremonies.

The Aha: MCP tool gates as live posture on the travel wall is not a binary you reboot. It is a function that should read live posture from Kiponos.io on every call. The host stays up. The leaf moves.

The problem: MCP tool gates as live posture on the travel wall lived in the process, not in the turn

Cursor is good at calling tools. It is not born with a shared, instant, restart-free control plane.

So teams hide MCP tool gates as live posture on the travel wall in the only places agent frameworks actually ship:

Where the gate hid	What you restart	What you lose
MCP server env / argv	The MCP process	Open tool sessions
Skill file on disk	The agent turn, sometimes the host	Context the model already paid for
Host-local JSON	Whatever still has the file open	Agreement between two agents
Hard-coded `if` on `tools-allow`	A release	The incident clock

The travel-coordinator wall already knew. Cursor did not, because it had started earlier.

That is the missing piece: the framework gave you tools. It did not give you a live hub.

What teams believe

Belief	Production
We'll catch it next turn	The travel-coordinator wall already knew this turn
Restart Cursor — it is cheap	Cheap until 14:12 ate the overbook thread the agent already paid for
The skill file is the source of truth	Skills instruct. They do not fan out
Put the SDK in the SPA	Connect tokens do not belong in a browser

The Aha: local get, live write, host stays up

Kiponos holds a nested tree. Java and Python SDKs keep the latest values in memory, patched over WebSocket deltas. The hot path inside a Cursor tool is a local get — no HTTP RTT per travel lookup.

Hub leaf for this essay:

examples/mcp-tool-gate-live/tools-allow = search,read

Runnable proof: examples/java/mcp-tool-gate-live

Public SDKs: Java, Python, plus React/Angular server peers (createFromEnv). Never put Connect tokens in the SPA.

Config tree (travel + peers)

examples/
  mcp-tool-gate-live/
    tools-allow: search,read          # MCP tool gates as live posture on the travel wall
apps/
  travel/
    live:
      tools-allow: search,read

Integration — Java hot path

Kiponos kip = Kiponos.createForCurrentTeam();
Folder gate = kip.getRootFolder()
        .folderOrCreate("examples")
        .folderOrCreate("mcp-tool-gate-live");
if (!gate.hasKey("tools-allow")) {
    gate.set("tools-allow", "search,read");
}
String posture = gate.get("tools-allow");
// Cursor tool: refuse the dangerous call when posture moved

Same leaf from a Python tool (Cursor just calls it):

from kiponos import Kiponos

k = Kiponos.connect(quiet=True)  # env: KIPONOS_ID, KIPONOS_ACCESS, KIPONOS
try:
    posture = k.get("examples/mcp-tool-gate-live/tools-allow", "search,read")
    if str(posture) == "search,read":
        raise PermissionError("MCP tool gates as live posture on the travel wall gated live — host not restarted")
finally:
    k.disconnect()

The Cursor process does not recycle. The next tool call already sees the dashboard edit.

Real scenarios

Event	Without Kiponos	With Kiponos
Group-chat flood during a gds blip	Restart Cursor; lose the overbook thread the agent already paid for	Set `tools-allow` live; next Cursor tool call already obeys
Peer host still on old tools-allow	Paste the value into the other chat	One hub leaf; both processes `get()` locally
travel-coordinator wall shows the new posture	Cursor started earlier so it writes anyway	Dashboard and tool share the same memory tree
Incident over, resume	Another Cursor restart	Set `tools-allow` back; session continues
Shopping admin still mutating stock mid-freeze	Two ceremonies, two lost threads	Same tree, two products, no paste

Performance (this path, not a generic table)

Cursor tool get() is an in-process map lookup after bootstrap.
One WebSocket per process lifetime — not per travel line.
A dashboard edit is a delta of tools-allow, not a config-file reload.
You do not pay model tokens to “please restart Cursor.”
A second host converges without a third paste onto shopping admin still mutating stock mid-freeze.

Compare to alternatives

Approach	Honest fit	Why it still restarts
Env file + Cursor reboot	Simple at 09:00	The freeze is at 14:12
Skill markdown as policy	Good instructions	Not a live bus
Redis poll inside the tool	Shared, but RTT on the hot path	You invented a hub with worse UX
Feature-flag SaaS	Product experiments	Rarely session-safe for Cursor
`@RefreshScope` / actuator	JVM apps	Does not restart Cursor

When not to use Kiponos

Situation	Why
Tool schema itself changed (new argument)	That is a code/Cursor restart
Secret rotation of Connect tokens	Credentials are not live knobs
One-off local script, no peers	A hub is overkill
Browser-only “SDK in the SPA”	Forbidden — tokens leak or defaults lie

Rehearsal beats slides

In staging: set a painful tools-allow, prove Cursor recovers without a host kill, prove clamps reject nonsense, prove last-known-good when the hub is firewalled. That drill ends half the architecture arguments about MCP tool gates as live posture on the travel wall.

Why Cursor is the wrong restart target

Cursor is good at calling tools. It is not a control plane. Killing it to flip tools-allow teaches the on-call that judgment requires a process ID. The travel-coordinator wall already disagrees.

What the travel operator actually said

At 14:12 someone said, out loud: mute writes on the noisy room — keep reads. That sentence is the whole product. If it cannot land in the running Cursor process in seconds, you do not have posture. You have a wiki.

Getting started (15 minutes)

TeamPro on kiponos.io → Connect → KIPONOS_ID / KIPONOS_ACCESS / profile ['my-app']['v1.0.0']['dev']['base'].
Clone github.com/kiponos-io/kiponos-io.
cd examples/java/mcp-tool-gate-live && cp kiponos.local.env.example kiponos.local.env
./gradlew test run — prints examples/mcp-tool-gate-live/tools-allow=...
In the dashboard, change tools-allow. Keep the process up. No rebuild.
Point your Cursor tool at the same leaf. Do not ship a new server binary to flip MCP tool gates as live posture on the travel wall.

The moral

If flipping MCP tool gates as live posture on the travel wall requires restarting Cursor, you do not have a gate. You have a hope with a process ID.

Agent frameworks already know how to call tools. Kiponos is the live hub they do not ship — so the travel-coordinator wall can change its mind without killing the session.

How to try: examples/java/mcp-tool-gate-live and ./gradlew test.

DEV Community