Your AI SaaS business, ready by this week!
TryΒ demoPublished Oct 5, 2026 β¦ 8 min read
Aikeedo v5.1 adds MCP connectors: chat can search Notion, read Sentry errors or call your own API, with each user signing in to their own account and approving every tool. Here is how it works for operators and subscribers.

A chat assistant that knows everything about the world and nothing about your user's own work only gets you so far. The question we kept hearing was a simple one: can the AI look at my Notion, my issues, my errors, my data?
In v5.1 the answer is yes. Connectors let the AI use tools from other services in the middle of a conversation β search a Notion workspace, look up a Linear issue, read a Sentry error, or call your own internal API β while your users stay in control of what it is allowed to do.
This is the biggest feature in the release, and this post covers what it is, how you run it as an operator, what your subscribers see, and where the limits are.
Connectors are built on the Model Context Protocol (MCP), the open standard that Claude and ChatGPT use for the same purpose. A service publishes an MCP server, and any app that speaks MCP can use the tools it offers. That means any remote MCP server works in Aikeedo, and the list of services grows without us writing an integration for each one.
We call them connectors, as Claude does, because that is what users understand. A user connects a service, and from then on the AI can use it in any of their chats β with or without one of your Assistants, and on any model that supports tool calling.
Two ideas shape the whole design:
Your subscribers manage connectors under Settings β Connectors, which has a Yours tab for what they have connected and a Discover tab for connectors you have made available.
Clicking Connect signs the user in to the service β most use OAuth β and then lists the tools the server offers. Tools are grouped by what the server says they do: read-only tools, write and delete tools, other tools, and tools that cannot be used here along with the reason why.
Every tool can be set to Always allow, Needs approval or Never allow, one at a time or for a whole group at once. Every tool starts at Needs approval, including the ones a server calls read-only. A server describes its own tools, and a server can get that description wrong or worse. We would rather ask once too often than let a destructive tool run because it was labelled harmless.
When the AI wants to use a tool that needs approval, the chat pauses and shows a card with the tool's name and exactly what it is about to send. The user chooses:
Closing the card declines the request and lets the user reply in their own words. Nothing hangs, and nothing runs without a decision.
When the plan allows it, any workspace member can add a server of their own. They paste its address, and a three-step flow takes over: it names the connector, checks the server live β reporting each step as it goes, such as connecting, finding the sign-in settings and checking the provider β and then lets the user confirm how it signs in. If the user's choice contradicts what the check found, they see a warning.
Custom connectors belong to the workspace, and each member connects with their own account. This is the route for your own APIs and for any MCP server that is not in a catalogue.
Connectors are off by default, and three switches must all be on before a user sees them.
0 to disallow custom connectors.This gives you a straightforward way to build plans around connectors. Offer a handful of curated services on an entry plan, the full catalogue on a higher one, and custom servers only where you are comfortable with them.
Operators also manage connectors in the admin panel: add a server by address, let Aikeedo check it, choose how users sign in, declare request headers such as an API key, and see the tools it offers. You can test a connector, refresh its tools on demand, and switch it off without deleting it.
There is no credit charge for connector calls. They are requests against the user's own account with the service, and the cost that matters β the model's tokens β is already metered. One thing to keep in mind: every tool offered to the model adds its description to the request, so connecting servers with a lot of tools makes each message larger. Aikeedo limits how many tools it offers per connector and per request, and picks the ones relevant to what the user is asking.
You can add servers by hand, but you should not have to. The new Connector Catalogue plugin ships 487 vetted MCP servers from official vendors, in 14 categories: code, productivity, data, sales and marketing, commerce, finance, communication, design, media, travel, education, legal, health and more.
Only first-party servers are included: each one has a namespace verified against the vendor's own domain, runs over HTTPS and was reachable when we built the catalogue. It is available in the marketplace, requires Aikeedo 5.1 or later, and you can add connectors yourself without it.
A connector receives whatever the AI sends it, and its tools act on a real account. We treated that seriously.
.env file, and are never sent back to the browser. Back the key up with your database: without it, every user has to reconnect.localhost, private networks or cloud metadata endpoints are refused, including through redirects. The same guard now protects web pages read during chat and dataset URLs.Update to v5.1, then:
APP_ENCRYPTION_KEY is set. Updates write it for you; if the file is not writable, run php bin/console app:generate-key.The Connectors documentation has the details, and the v5.1 release post covers everything else that is new.
Try the live demo to see it in action, or head to our pricing page if you are ready to get started.
Discover more on related topics that might interest you