The MCP gateway between your accounts and your agents

Connect once.
Every AI agent gets exactly the right tools.

Connect work and personal accounts once. Claude, Claude Code, Codex, Cursor, Hermes, and any MCP client reuse them through one URL — each with its own rules. No re-auth. No over-sharing.

Connect your first account
and any MCP client
tap an agent ↓
Access map Claude Code — both Sentries, work Linear, analytics
AIP
Sentry Work account allow
Sentry Personal account allow
Linear Work · 14 tools confirm
Notion Personal · 9 tools hidden
PostHog Work · 8 tools allow
4 of 5 connections exposed  · same connections, reused
Linear Notion PostHog Sentry Exa Search Granola Fireflies X Paddle Numon Atlassian Rovo Jira Confluence Cloudflare Supabase DigitalOcean Hevy WHOOP Google Maps Google Calendar Google Sheets Gmail Yandex Direct Things 3 Apple Calendar Linear Notion PostHog Sentry Exa Search Granola Fireflies X Paddle Numon Atlassian Rovo Jira Confluence Cloudflare Supabase DigitalOcean Hevy WHOOP Google Maps Google Calendar Google Sheets Gmail Yandex Direct Things 3 Apple Calendar
Problem 01

Cursor is banned at work, Claude Code isn’t — my Linear token can’t tell the difference.

Per tool. Per account. Per agent.

Not one big OAuth grant. Claude Code gets both Linears. Cursor gets the personal one — the work one isn’t even in its tool list. Every cell is a rule you set once: each agent gets its own view of your tools, and what’s hidden simply isn’t there.

Hide — the agent can't even see the tool exists.
Allow — safe reads run instantly, no nagging.
Confirm — destructive stuff waits for your tap.
it's live — tap the cells
Linear for agent Claude Code
create_issue W
work
personal
list_issues R
work
personal
update_issue W
work
personal
2 of 3 tools hidden on the personal account · 2 need your confirmation.
Google Calendar 2 accounts
P Personal you@gmail.com allow
W Work you@company.com confirm
Add another account
Claude sees two calendars. You manage one connection.
Problem 02

Claude connects one Google Calendar. I have two. It double-booked me within a week.

Add the second account. And the third.

One connection, any number of accounts inside it. Each account keeps its own rules per agent — Claude finally sees both calendars, and the work one still plays by work rules. To an agent, they’re just two separate calendars.

Problem 03

Our billing system has never heard of MCP. The vendor’s roadmap says 2027.

One permission layer.
Three ways in.

No MCP required on the other side. The catalog wraps plain APIs — OAuth, API keys, OpenAPI specs — and anything that already speaks MCP drops in by URL. Out comes MCP with OAuth DCR, the way Claude connects natively, or a key-protected OpenAPI for the agent you wrote yourself.

Built-in chat

Test in AIP Chat

Talk to your connected tools right in the dashboard. Approve or reject risky calls inline before wiring anything up.

dash.aipx.run/chat
MCP agent

A scoped MCP endpoint

One URL per agent. Claude, Claude Code, Codex, Cursor — anything that speaks MCP gets only the tools you flipped on.

https://dash.aipx.run/mcp/:agent
OpenAPI agent

A key-protected API

A generated OpenAPI document plus an agent key. Call approved tools from Hermes or any backend, no MCP required.

GET /api/agents/:agent/openapi
Use with ClaudeClaude CodeCodexCursorHermesYour own backend
Problem 04

Gmail says I emailed my boss at 2am. I was asleep. One of my agents wasn’t.

Ask the log, not the agents.

Agent, tool, account, inputs, result, timestamp — every call your agents make, on the record. When something looks off, you don’t interrogate five chat histories. You look up the row and know exactly which agent to ground.

Audit log every agent · every call
14:02 Claude Code list_issues success
13:47 Claude create_event success
13:12 Codex update_issue error
02:14 Hermes send_message success
the 2am email, found
Input 
{ "to": "boss@company.com", "subject": "my honest feedback" }
Result 
{ "id": "msg_18f4c2", "labelIds": ["SENT"] }
02:14 · Hermes · the work Gmail. Case closed — hide send_message from Hermes and go back to sleep.

Three steps. Then you never think about it again.

1

Connect accounts

OAuth into Linear, Sentry, Notion, PostHog — or paste any MCP URL. Work, personal, side project: as many accounts per integration as you live in.

2

Scope every agent

The same connections, reused by every agent — but each one sees only what you flip on. Change your mind? Flip it back. Nothing re-authenticates.

3

Paste one URL

Each agent is a scoped MCP endpoint or a key-protected OpenAPI. Drop it into Claude, Claude Code, Codex, Hermes — or your own backend.

Pricing

Simple. Counted in accounts, not in seats.

Tool calls and audit are unlimited on every plan. Upgrade when you need more agents, accounts, or integrations that work without authorization.

Free
$0 forever
  • Up to 3 connected accounts
  • 1 agent, unlimited tool calls
  • Granular permissions and audit log
Start free
Pro
$3 / month

Taxes included — we cover them.

  • Unlimited connected accounts and agents
  • Pro integrations
  • Cancel anytime
Upgrade to Pro

Stop re-connecting
the same five tools.

Connect them once, scope your agents, and get back to work. Start free with 3 connected accounts.

Open the dashboard