CypherMonkeyAI omni-channel lab

A lifelong-learner lab on AI omni-channel agents, including voice and the contact center. Free to read. Nothing sold.

Model Context Protocol · L2 · 20 min

Consent and roots

Learn in five minutes

  • The person should see which server and which tool are about to run.
  • A root is a boundary on which files a server may touch. Confirm the spec's current word for that boundary.
  • A server that can read every file is a different product from a server that can read one folder.
  • Approval for a draft is still required before a send.
  • OWASP's LLM risks include excessive agency. Read the current list.

Grant the notes folder, not the home folder. The trace records the folder. A later reviewer can see the boundary.

This section HTML

Visual

T6-R5 diagram

Diagram SVGDiagram PNGDiagram HTML

One folder

  1. Write the allowed path as notes/ only.
  2. A request for ../secrets fails the check.
  3. The trace stores the denial.
  4. No file outside notes/ is read in the story.

This section HTML

Quiz

What is the point of a root or boundary?

The boundary is the permission.

What does excessive agency mean at learning depth?

OWASP uses that family of risk.

What does the trace store in the example?

The denial is the evidence.

This section HTML

Ecosystem

Root
A boundary on files or data a server may see. Confirm the spec's word.

Sisters: Plugins, Function calling, Language-server ideas.

A notes server can serve chat and a voice host if both speak the same protocol.

Treat the spec revision as something you pin, like a model id.

Config

  • Write the goal in one sentence.
  • Name the input and the allowed output.
  • Save the trace next to the result.

Troubleshooting

  • The answer looks fluent and is wrong. Check whether a source passage actually contains the claim.
  • The loop never stops. Set a step cap and stop when the tool result repeats.

This section HTML

Sources HTML

Sources and go deeper

sources.json