Skip to content
AI & ChatOps · for operators

Operate from where you already work

Drive ispforge with an AI assistant over MCP, or run day-to-day peering and alerts straight from Slack and Discord — every action through the same risk scoring, approval gates and audit trail as the portal. The AI proposes; a human still approves.

AI · MCP

Describe the intent — ispforge does the rest, safely

ispforge runs a Model Context Protocol server with 50+ tools across read, write and lifecycle. Point Claude, Claude Code or any MCP client at it and the assistant can investigate your network and propose changes in plain language — you don't have to remember community values, prefix-list names, or vendor syntax. Every write becomes an intent: risk-scored, previewed as a diff, and gated on approval.

ispforge mcp — drain a peer
» "drain my Cogent transit at edge1.nyc for 24h — they've got planned maintenance" intent bgp.apply_te_overlay · drain AS174 action graceful-shutdown + local-pref 0 risk high · approval required state pending_approval # nothing is on the router yet — a human approves first
The intent loop

The AI proposes. A human approves. Everything is audited.

An AI-driven change takes exactly the path a human takes through the portal — kicked off in plain language instead of clicks.

01 · ASK

Plain language

"Depref my Hetzner peer at edge1.lon by 50."

02 · PLAN

One MCP verb

Resolved to a single tool call with parameters.

03 · SCORE

Risk low / med / high

Blast radius and customer impact assessed.

04 · VALIDATE

Same gates as the portal

Subnet, RPKI, max-prefix, your approval policy.

05 · APPROVE

Human signs off

Review the real diff; two-person rule optional.

06 · EXECUTE

Agent deploys, audited

Commit-confirmed, rollback armed, four audit rows.

ispforge mcp — refused at validate
» "peer with AS6939 at SFMIX, peer IP 10.0.0.1" intent bgp.create_peering_session validate ✗ subnet_membership 10.0.0.1 not in SFMIX LAN 206.197.187.0/24 state failed · never reached the router
Guardrails

The AI can't bypass policy

The intent pipeline runs the same validation an operator hits in the portal. A bad peer IP, an RPKI-invalid origin, a max-prefix out of bounds — refused at validate, never reaching a router. On top of that you scope each assistant:

  • Per-key intent allow-list — this assistant may only propose the verbs you permit (e.g. TE overlays, not member invites).
  • Blast-radius capmax_affected_targets bounds how many sessions or routers one intent can touch.
  • Two-person rule — require a different approver from the proposer, enforced for cookie, user-JWT and API-key auth alike.
  • Full audit chain — proposer → approver → worker → outcome, one row each, on every change.
Risk-scored by default

Routine work flows; risky work waits for you

LOW · auto

Runs immediately

Preview a router, refresh an IRR prefix set, acknowledge or resolve drift, sync from PeeringDB. Self-executes if you allow low-risk auto-approve — no round trip.

MEDIUM · approve

Needs a human

Deploy a router, assign a policy, accept or reject a peering request, edit inventory. Proposed, previewed, and held until someone approves.

HIGH · always approve

Customer-affecting

Drain or disable a session, create a peering session. Always requires approval — a depref against a customer is rejected outright.

MCP access is included on the ISP plan and up

Slack & Discord

Run peering from the chat you already live in

Native Slack and Discord bots — installed with OAuth in two clicks, no legacy webhooks. They post rich, formatted alerts and link straight to the right page in ispforge. From Discord, accept or decline peering requests on interactive buttons without leaving the channel. Prefer one-way? Telegram and email endpoints too.

#noc-alerts
i
ISPForge APP 10:24
New peering request · AS13335 · Cloudflare
Verified via PeeringDB · SFMIX, DE-CIX New York
Accept Decline View details
// PEERING

Requests, in the channel

Inbound requests arrive with the peer's verified details; accept, decline, or open the request — all from chat. Accepts and declines notify the peer automatically.

// OPS

Deploys & drift

Deployment success and failure, BGP drift detected, and session state changes land in your NOC channel with a View → link to the exact page.

// REACH

Wherever your team is

Slack, Discord, Telegram and email endpoints, each installed independently per org. Send a test alert on setup to confirm delivery.

Operate your way

Portal, CLI, AI, or chat — same guardrails

Every path runs through the same intent pipeline and audit trail. Join the beta, wire up an assistant or a chat bot, and drive your network however you like.