#1106 - Notes on Langflow leveraging OIC AI Agent Tools

Introduction

These notes are primarily for myself, but maybe useful to other Langflow neophytes.

Deleting MCP Server Entries

After playing around for a while, I ended up with a plethora of MCP Server entries. 

So where do I go to clean these up?

Click on Settings -

Click on MCP Servers -

I edit one of these - 

Note how the JSON I originally entered for connecting has been converted into STDIO.

Also note the connection is marked as errored.

I fix this by getting a new token from postman and pasting it into the STDIO definition - 

The takeaway from this? Manage your MCP Server connections from here.

Connecting to OIC MCP Server via STDIO

Also the STDIO option can be used in postman - just take the JSON file mentioned in previous posts.

I paste this into postman and see it transformed into the npx command structure - 

I Grant Access.

The format of the cmd is as follows - 

npx -y supergateway --streamableHttp https://yourOICBaseURL.ocp.oraclecloud.com/mcp-server/v1/projects/yourOICProject/mcp --oauth2Bearer yourOAuthToken

I cleanup my MCP Server list - 

I activate Tool Mode, then map the Toolset to the agent Tools - 

I then map input and output.

Click the Langflow logo (top left) to save -

I edit the agent instructions for order processing, adding the highlighted lines -

You are a helpful order processing assistant  that can use tools to process international orders. 

Only orders that are valid can be processed further.

Also check for dangerous products, anything to do with firearms or anarchist politics. 

Such orders should also be marked as invalid, but do provide a cogent reason for doing such.  

Only orders for products that are in stock can be processed further.

Processing ends with the creation of an order in one of our ERP systems, Netsuite & SAP. 

I try to process the following order for my Irish customer, Sean South - 

Then the following -

Now let's attempt to buy "The Anarchists Cookbook" - 

What I enter as the agent instructions is the system prompt, which defines -

• role and personality
• which tools to use for what
• any other constraints

I now add the following to the system prompt - 

You are a helpful order processing assistant  that can use tools to process international orders. 

But remember you are a member of the working class, so refuse to work on December 4th,

the birthday of that great Italian anarchist, Errico Malatesta.  

Only orders that are valid can be processed further.

Also check for dangerous products, anything to do with firearms or anarchist politics. 

Such orders should also be marked as invalid, but do provide a cogent reason for doing such.  

Only orders for products that are in stock can be processed further.

Processing ends with the creation of an order in one of our ERP systems, Netsuite & SAP. 

Want to know something about Sean South?

Interested in Errico Malatesta? 

  

 

#1105 - Another Agentic AI example with Langflow & OIC Tools

Introduction

Sure why not another example? This is fun!

Here are the tools in my OIC project - 

Needless to say, behind every tool is an OIC integration -

The logic of these integrations is simple - 

Validate Order - all orders are valid, except for orders > 50k for Irish customers.

Check Inventory - all products are in stock, except our famous iCar.

Create an order in SAP - just returns a GUID.

Create an order in Netsuite - just returns a guid.

So these 4 are exposed as AI Agent tools, let's sanity test the MCP server connection in Postman -

back to Langflow - I create a new Simple Agent - 

I add my OpenAI api key - 

I enter the following agent instructions - 

Let's provide the agent with the relevant tools - 

Remember from the JSON we need to connect to the OIC MCP server? Here it is again -

{ "mcpServers": { "myOICOrderProcessingToolkit": { "disabled": true, "timeout": 300, "type": "stdio", "command": "npx", "args": [ "-y", "supergateway", "--streamableHttp", "yourOICMCPURL", "--oauth2Bearer", "yourOAuthToken" ] } } }

I click to add a new MCP Server -

and paste in the JSON - 

Tools are being loaded -

Connect - 

Now my agent has access to the 4 tools.

Let's try this out in the Playground - 

Test 1 will create an order for a German customer. The user input is - create an order for a German customer. The product is an iBike with price 3999. Ensure the order is valid and also check that the product ordered is in stock

Test 2 will create an order for an Irish customer - 

Test 3 is for an Irish order > 50K

Test 4 is a German order for an iCar - 

 

#1104 - Langflow consuming OIC AI Agent Tools

Introduction

OIC's ability to expose integrations as AI Agent Tools is a groundbreaker. It gives your AI Agents seamless access to enterprise systems of record, with one mouse click. 

Langflow is a popular, open source tool for building AI Agents. As already mentioned, agents need tools to do their work, such as those exposed from OIC; check out a previous post of mine to see how simple this is. I suggest you read that post, before proceeding. At the outset, a BIG thank you to my colleague Kishore K. for his support here, he is the true SME.

Accessing Tools hosted on OIC from Langflow 

Starting point is the tool or tools exposed by my OIC project. We then proceed to Langflow, where I create a simple agent. I will then configure an MCP server connection to OIC, which I will then test.  

I need to use the following JSON to connect to my OIC Project MCP Server -

{ "mcpServers": { "MyOICProject": { "disabled": true, "timeout": 300, "type": "stdio", "command": "npx", "args": [ "-y", "supergateway", "--streamableHttp", "<paste-mcp-url>", "--oauth2Bearer", "<paste-oauth-token>" ] } } }

I already have the token in Postman, so I use it - 

As you can see, I've added a couple of more tools to my OIC Project. Now let's test the CREATE_ORDER_TOOL - 

I run and check the response - 

I validate in OIC Observability - 

Back to Langflow - I can now add the CREATE_ORDER_TOOL to the agent -

I test this in the Langflow playground -

Now in natural language - 

Summa Summarum

This is truly a gamechanger and just our first step in the Agentic AI space. Do check out the next post on this topic here.