[IN DEV] Fibery MCP server

[aoe]

Hi,

Has the Fibery team discussed the possibility of creating an MCP server as a bridge for interacting with the API?

From a personal point of view, I’ve been trying different MCP servers through Claude Desktop as host and when you combine multiple of them and their respective tools, it almost feels like magic.

One example scenario:

• Research competitors (via Brave Search MCP server)
• Summarize and analyze with Claude
• Create Task in Fibery to read analysis later with Due Date this week (via Fibery MCP server)

This would all be done in an MCP Host (e.g. Claude Desktop).

There are servers for e.g. Airtable/Atlassian but they are not official implementations. They are however larger in size and more known so their users have created these implementations. I guess there’s nothing stopping me or anyone else from trying to create one themselves but it would be interesting to know if it is in consideration. It could of course potentially attract new customers as well.

References:

• Introduction - Model Context Protocol
• GitHub - modelcontextprotocol/servers: Model Context Protocol Servers

[mdubakov]

Yes, we are thinking about it

What are your imaginary use cases for Fibery, what do you think?

[aoe]

Is it ok to write…anything and everything?

For me, this is a case of “provide the opportunity and the solutions will come”.

An example: I just wrote an MCP server with the help of Claude 3.5 (and some debugging) in a few hours that can interact with our financial SaaS through their REST API. The server provides just one tool for now that retrieves all info about the customer (credit rating, invoiced amount, unpaid invoices etc.).

Now instead of going through a clunky graphical interface, I can ask about this information using natural language through a chat interface. Extrapolating, we can create invoices through chat, look at receipts etc. There’s too much to list here. More importantly, other users without coding knowledge can do the same thing (not necessarily a good thing here concerning financial data but I think you get the point :)). And they don’t have to learn a new interface just because it’s a new SaaS (functionality may be good but lacking in UI).

Now using the same analogy with Fibery, basically having the choice to interact with the databases through chat would be huge. And I can come up with multiple use cases (as I’m sure you can as well) but I thought I could provide more value if I explain my thoughts at this point.

An idea that I had (most likely naive) would be to lookup the schema for different databases through the REST API and then produce queries for that schema (since everyone’s workspace would be different).

Now one specific (and trivial example):

• I want to get financial data through our SaaS as mentioned earlier for Customer A → I ask through chat to retrieve the Customer in fin-SaaS
• I already have a Customer database in Fibery with similar information, I want to make sure the information is up to date or just compare the information → I ask through chat to retrieve the Customer in Fibery
• I can manually compare, if something is wrong, ask to update either service through chat.

I hope it gives an idea at least. I’m still very early in my learning journey and there’s a lot I don’t know (wish I could allocate more time but don’t we all :))

[derbenoo]

I looked at the MCP protocol over the weekend and build an MCP Server for Fibery

Claude understands that I want to create a support ticket entity inside my Fibery workspace:

It works! The ticket was created:

Here is a sharelink to the example MCP Tool I build:
https://fiberflow.fiberflow.io/@public/Templates/Ticket/470890d1-9d28-4b0c-9846-6f22a5512eff

I started out with generic tools like “Get the schema of a Fibery database” and “Create an entity” (this is how the Airtable MCP works). I quickly realized that this is very difficult for the AI to get right. The AI Agent has to understand the Fibery schema and how each field is structured in order to create an entity or query a database with the correct select statement.

Instead, the user has to be able to build their own MCP Server Tools for their Fibery setup. Fibery is a flexible platform where the user can build whatever they want, so the MCP Server has to be very flexible as well. The user should be able to craft dedicated MCP Tools that allow the AI Agent to perform certain use-cases inside the Fibery workspace. A list of useful things that the AI can do with a lot of context provided via the description of the respective tool. This is also also a lot safer than to give the AI full access to every database and allow read, create, update and delete operation on every entity in Fibery

I also discovered that the response the MCP Tool sends back to the AI Agent has to be very flexible as well. The formatting of the returned entity might be better as a well-crafted plaintext that only contains the relevant properties instead of just JSON.stringify the entire entity. Also if an error occurs, the AI has to exactly know what went wrong, why and how to fix the issue.

So here is the solution I came up with: The user can build an MCP Tool with a no-code interface by providing the inputs of the tool (what the AI has to provide when calling the tool), the actions that should be performed inside the Fibery workspace (e.g., create a new support ticket) and what to respond back to the AI Agent. So each MCP Tool is an automation that can be executed by the AI Agent.

Testing the MCP Tool by asking the AI Agent to call it every time is very cumbersome, so I added the option to manually run the MCP Tool:

The MCP server allows for inputs to be required / optional, I added that option but the Claude AI Agent does not seem to understand that an input is optional at this moment

Here is a simple example where I use the flexibility of the automation engine to give the AI Agent a tool that solves a use-case instead of only doing basic data manipulation. I will add more actions in the next couple of days like “Search for entities” action with filtering, which is very useful for building MCP Tools.

I am curious if this is way of building an MCP Server for Fibery is useful to you or whether you had something totally different in mind You can try it out over at app.fiberflow.io

Cheers,
Ben

[derbenoo]

Added some documentation how to add the Fiberflow MCP Server to Claude Desktop:
Fiberflow - AI agent called trigger

[aoe]

Looks nice and well thought out, will take a look later!

[maxdubakov]

Hey!

After seeing all the discussion here, we went ahead and built an official Fibery MCP Server - it’s now live!

You can now chat with your Fibery workspace through Claude or any other MCP-supporting LLM. Current tools enable LLM to check schema, query databases, create and update entities.

If you want to try it out, check out our repo for the setup details: GitHub - Fibery-inc/fibery-mcp-server

Would love to hear what you think if you give it a shot

[Renato_Carvalho]

I just tested over here! And, wow!

[aoe]

Will try to test this as soon as possible, looks very good!

Would this still be the way forward if you want to limit permissions? API Key Permissions - #2 by Chr1sG

[mdubakov]

One real use case

Check Linear competitor entity in Fibery and see what information is missing. Find relevant information in the internet and update Linear entity in Fibery with this information.

Result

Fibery

[RonMakesSystems]

Hey!

This looks really cool! Does this give AI have access to photos in the files field? As well photos in rich text embed?