The rise of multi-step and multi-agent orchestration
As organizations move beyond single, stand-alone AI assistants toward multi-agent systems, the complexity of integrating data, tools, and enterprise AI workflows grows exponentially.
In these architectures, each agent might specialize — one in analytics, another in customer engagement, another in data quality — yet they must collaborate seamlessly and securely.
That orchestration layer is where Model Context Protocol (MCP) enters the scene. MCP provides a standardized way for agents to discover, authenticate, and call tools or data sources. Instead of every agent hardcoding access to APIs or databases, MCP servers expose those capabilities in a consistent, declarative way.
But as soon as teams begin to scale beyond a few experiments, a new challenge emerges — managing multiple MCP servers and its associated governance.
The challenge: Managing multiple MCP servers
While MCP simplifies interoperability, it introduces management overhead as organizations and enterprise AI projects scale:
- Each MCP server has its own configuration, credentials, and permissions.
- Teams must maintain consistent schemas and metadata across environments.
- Monitoring, rate limiting, and audit logging need to be centralized to ensure compliance.
- Upgrades or policy changes across multiple MCP instances can quickly become operationally heavy.
Without centralization, developers end up maintaining dozens of individual MCP deployments — each potentially diverging from enterprise standards or security policies.
Vendia’s fully managed MCP-as-a-service works like a SaaS application. We handle deployments, availability, load balancing, tracking the MCP standard and all the rest of the hard stuff – from unified registration & discovery of your APIs and MCP servers to centralized policy enforcement with RBAC & data masking and provide end to end observability. Developers can focus on building AI agents that do anything from claims processing for an insurance company to image analysis to summarizing reports without maintaining protocol infrastructure or the underlying plumbing while maintaining governance and control.
Tutorial: Multi-step workflow orchestration with Vendia’s MCP solution
Imagine a multi-step orchestration scenario in an enterprise to automate data processing, report creation and report notification across Snowflake, Amazon S3 and Slack respectively. In order to accomplish this, a developer would need to configure three different MCP servers, one for each data source or technology.
With Vendia, all of this can be accomplished with minimal overhead and management. Vendia enables developers, data analysts and administrators to connect Vendia’s fully managed MCP server to their backend data sources like Amazon S3, SaaS platforms like Slack, Hubspot, Snowflake and expose their APIs as MCP tools which can be accessed via MCP compliant AI agents (Claude, ChatGPT etc.).
The goal would be to create a coordinated workflow across a single AI agent.
- Data processing step: Reads data from Snowflake, creates a report and uploads it to Amazon S3.
- Notifications step: Notifies a user or a channel via Slack and shares the details of the summarized report.
To get started, we will create a storage connection and configure a couple of APIs in Vendia, followed by connecting an AI agent to Vendia.
Storage Connections
Vendia enables connecting to Amazon S3 via the Storage connections capability. Storage connections bridge the gap between your existing data infrastructure like S3 and your AI application. With the Amazon S3 storage connection, AI agents can connect to your S3 buckets and perform file operations (browse, read, write) at scale. In order for an AI agent to get access to your S3 bucket(s), your AI ops or IT admin needs to configure IAM permissions and Access policies to govern which files and folders the AI agent can access.
In this example, we already have a storage connection created in Vendia called Marketing-finance-reports. Within this connection we have a bucket called vendia-client-files-testing-bucket which has a collection of files (.mp4, .txt, .html, .md, .pdf etc.) and folders already created prior to connecting the bucket to Vendia.
We also have created a folder via the S3 console called marketing analysis reports within this bucket.
API Catalog
The Vendia API catalog enables AI agents to access your enterprise APIs – internal (private), external or public (third-party). The API catalog currently supports OpenAPI/Swagger specifications in .json, .yaml and .yml formats, which are made available as MCP tools for AI clients to consume.
- Slack API: For the slack API, we have downloaded the swagger specification from https://github.com/slackapi/slack-api-specs/blob/master/web-api/slack_web_openapi_v2.json
In this example, we have created a Slack ‘app’ in the Vendia Slack workspace, that
generated a bot token for the app. You can look for the bot token under OAuth & Permissions in your slack ‘app’ console as shown below:
Additionally, the app is configured with specific scopes which governs what operations the token has access to as per the below image:
The screenshot below shows the parameters required to configure a Slack
specification.
Similarly, we have an API catalog configured for Snowflake.
- Snowflake API: For the Snowflake API, we got the API specification from https://github.com/snowflakedb/snowflake-rest-api-specs/blob/main/specifications/sqlapi.yaml
Edit the ‘url’ field with the Snowflake URL of your account as shown in the screenshot below.
Next, you need to generate a Programmatic Access Token (PAT) for Snowflake. In
order to do that, follow the prerequisites , understand the privileges required and steps
to generate the PAT.
Once the PAT is generated, you can add it to the header section in the API catalog
configuration.
Connecting and using your AI agent
Once the storage connection and API catalogs are configured in Vendia, we need to connect our AI agent to Vendia’s MCP server. In this example, we have Claude (you can use any MCP compatible AI client) connected to Vendia via a connector as shown below:
Once our connection is set up, we are ready to orchestrate a multi-step workflow across Snowflake, Amazon S3 and Slack via Claude.
To begin with, we will list all the APIs and storage connections available to use.
Claude returns the APIs that have been currently configured in my Vendia account. This is exactly what we see in the Vendia dashboard.
Next, we will ask Claude to list storage connections available in Vendia. As you can see, Claude returns all the available storage connections.
Similarly, in our Vendia dashboard we have 3 storage connections as shown below:
Next up, we will query the Snowflake and Slack APIs from the API catalog. In the below prompt, I am asking Claude to list all the tables available in the Snowflake instance. First up, Claude explores the API catalog to see what operations are available, followed by executing a submit statement (POST operation) that executes a SQL query (SHOW TABLES) that lists tables.
In a nutshell, Claude translates the prompt into appropriate MCP tools that are available via the API specification while ensuring that operations can be executed based on permissions and allowed scopes.
Next I asked Claude to run its magic and help us uncover what insights we could get from the AA_TRAVEL_2209 table.
Claude came up with the below list of possible analytical outcomes we could uncover.
Things look great so far and now we’re perfectly positioned to orchestrate a multi-step workflow across the three systems – Snowflake, Amazon S3 and Slack. I finally asked Claude to generate a brand penetration report, followed by uploading the report in an S3 bucket and subsequently sending the report summary to a slack channel in the Vendia Slack workspace called ‘AI’.
As you can see below, Claude is working hand-in-glove with Vendia to list storage connections, invoking the permissible set of APIs across all the 3 systems.
In a few minutes, Voila! Claude has executed a multi-step workflow with a single prompt!
As per the prompt, Claude created a brand affinity report in Snowflake via the API operations allowed in the API specification and wrote the analysis to an html file which was uploaded to the specified folder (test-uploads) in the specified storage connection (matt-test-bucket). Vendia does all the heavy lifting of reading and writing to S3 buckets via the Storage connection.
Once the file was uploaded to the S3 bucket, Claude invoked the Slack-Vendia-MCP API via Vendia and was able to send the report summary to the AI channel based on the scopes allowed.
MCP and AI Governance
To recap, as an AIOps admin, I was able to set up permissions across all my systems within a couple of hours and as a business user I was able to generate my business outcomes (analytical insights + multi-step workflow execution) within minutes using Vendia and Claude. If I had to set up and execute a multi-step workflow without Vendia – I would have had to connect to individual MCP servers, configure them (some systems might not even have an MCP server) and manage each of them individually. As the number of backend systems and API integration grow over time with MCP – this exposes security and compliance vulnerabilities with increased AI tool usage.
Vendia’s core philosophy is built on the foundation of enterprise governance. Understanding how your data is used is more important than ever once an LLM is in the picture. With our new Read and Write Receipts for Data, Vendia records every data interaction by AI agents on an immutable ledger. This provides a complete, tamper-proof audit trail that strengthens your governance and compliance reporting. You gain full visibility into every data interaction by downstream applications, like AI and analytics workloads, giving you an unprecedented level of transparency and control.
To demonstrate Vendia’s value proposition related to MCP and AI governance, let us walk through some scenarios supported in Vendia.
- Disabling an API: In the screenshot below, we have disabled the Slack-Vendia-MCP API from the API catalog. Upon disabling the API, Vendia makes sure if any AI client is requesting access to that API will not be able to via security and governance controls. The desired behavior via the AI client, Claude is in this case should be to not be able to access the Slack-Vendia-MCP API.
Vendia honors that the API is disabled and accordingly does not allow Claude to access it. From an observability standpoint, Vendia logs all AI client calls and any external system access. For this example, Vendia logs this call as a ‘Read receipt’ of type GraphQL query in its UI as shown below.
- Editing S3 bucket access policy: Another powerful capability of Vendia is configuring S3 bucket access policies. In this example, we are updating the access policy for the test-uploads folder in the matt-test-bucket storage configuration. We updated the access policy to not allow AI agents creating or writing any new files to the test-uploads folder as shown in the screenshot below. Disabling these policies means that Claude (in our example) should not write or create any new files in this location.
Next, we execute a prompt from Claude to re-upload the previously generated brand affinity report to the marketing analysis reports folder in the marketing-finance-reports storage connection. As you can see below, Claude reports back saying it cannot perform the operation due to access restrictions – proving that Vendia does not allow this AI agent to create or write any new file to the specified location!
Additionally, in the Observe tab, we can see the read receipt for the ‘file create’ operation as failed.
On hovering on the ‘file create’ column you can further view which resource or file was being attempted to be created or written by Claude.
These are just a couple of examples on how Vendia enforces MCP governance. We will do a deep dive tutorial on other use cases related to MCP governance with Vendia.
Ready to give it a try? Sign up for free and turn your Amazon S3, Snowflake or other data source into AI outcomes in minutes. Contact our sales team at [email protected] for the Pro or Enterprise tier.