26 — AI Agents and MCP
| Owner | CSI Technology & Documentation Lead |
|---|---|
| Version | 1.0 |
| Status | Under Review — CSI MCP Gateway v0.1.4, n8n and Telegram operations are technically validated; proactive alerting and extended pilot remain pending |
| Review frequency | Quarterly and after material change |
| Applicable environments | CSI Lab, Development, Testing, Staging, Production, Disaster Recovery, Client Site |
Required coverage
Include:
- AI agents
- Tool calling
- Agent memory
- Short-term memory
- Long-term memory
- Planning
- Task execution
- Supervisor agent
- Worker agents
- Multi-agent systems
- Human approval
- Guardrails
- Permissions
- Sandboxing
- Audit logs
- Agent identity
- Agent credentials
- Agent limits
- Error recovery
- Agent monitoring
- MCP clients
- MCP servers
- MCP tools
- MCP resources
- Local MCP
- Cloud MCP
- Security
- Data access
- Approval controls
Create:
- Agent architecture
- Agent permission matrix
- Tool register
- MCP server register
- Agent testing checklist
- Agent risk checklist
Acceptance rule
This chapter becomes Active only after accuracy review, command or workflow testing where applicable, risk review, owner assignment, approval, and change-history entry.
CSI MCP Gateway — official reference order
- Build, deployment, operations and recovery: CSI-RUN-MCP-001 — CSI MCP Gateway v0.1.4 Build, Deployment & Recovery Runbook
- Source files, configuration and dependency lock: CSI-MCP-FILE-001 — CSI MCP Gateway v0.1.4 Source Files & Configuration Register
- Validation evidence, known errors and pilot record: CSI-MCP-VAL-001 — Live Validation, Troubleshooting & Pilot Record — 2026-07-28
Current verified endpoint: https://mcp.cyberspaceinfocom.com/mcp
Current verified health endpoint: https://mcp.cyberspaceinfocom.com/healthz
Current verified release: 0.1.4
Related pages
Change history
| Date | Version | Change |
|---|---|---|
| 28 July 2026 | 1.0 | Initial chapter brief created from the master framework. |
| 28 July 2026 | 1.1 | Added the verified CSI MCP Gateway v0.1.3 runbook, source-file register, validation evidence and future-support reference order. |
CSI-RUN-MCP-001 — CSI MCP Gateway v0.1.4 Build, Deployment & Recovery Runbook
CSI-MCP-FILE-001 — CSI MCP Gateway v0.1.4 Source Files & Configuration Register
CSI-MCP-VAL-001 — Live Validation, Troubleshooting & Pilot Record — 2026-07-28
CSI-RUN-MCP-N8N-001 — n8n ↔ CSI MCP Gateway v0.1.4 Integration Runbook
Verified n8n integration — 29 July 2026
- n8n integration, monitoring extension and recovery: CSI-RUN-MCP-N8N-001 — n8n ↔ CSI MCP Gateway v0.1.4 Integration Runbook
Current verified release: 0.1.4
Current verified automated tests: 6 files; 18 tests passed.
Current direct ChatGPT constraint: custom MCP app availability depends on eligible ChatGPT workspace entitlement.
Telegram operations channel and monitoring expansion — 29 July 2026
Current verified targets: Linux server, OPNsense, Grafana, Nextcloud, n8n, MCP Gateway, Prometheus, Blackbox Exporter and Nginx Proxy Manager.
Security boundaries remain unchanged: no Docker socket in MCP, no shell tool, explicit service allow-list, encrypted n8n credentials and read-only MCP operations.
Next automation sequence
- Completed: exact
/statusand approved operator Chat-ID gate; unrelated messages are silently ignored. - Next: build service-down plus recovery alerts with duplicate suppression.
- Build daily scheduled health summary.
- Add ERPNext, Portainer, Homepage and RustDesk monitoring.
- Add database, Redis, queue and runner health.
- Complete seven-day pilot and rollback drill.
- Begin n8n AI Agent integration only after monitoring controls stabilize.
- Defer voice control until AI-agent authorization and approval controls are validated.
Detailed evidence and procedures: CSI-RUN-MCP-N8N-001 — n8n ↔ CSI MCP Gateway v0.1.4 Integration Runbook
Operational status update — 3 August 2026
- [x] CSI MCP Gateway deployed and healthy
- [x] Authenticated n8n-to-MCP call validated
- [x] Telegram status workflow published
- [x] Approved operator gate validated
- [x] Blackbox probe data included in the service-health path
- [x] Read-only security boundaries preserved
- [ ] Proactive alert/recovery automation
- [ ] Seven-day pilot
- [ ] Client-impacting agent actions; prohibited until a separate approval and liability framework is approved
The CSI operating policy remains: agents may assist, analyse and report, but must not make uncontrolled changes inside client infrastructure.