MCP Servers

MCP Servers act as a standardized bridge between AI models and real-world systems. They expose tools, resources, and prompts in a structured and secure way, allowing AI agents to interact with databases, internal services, or external endpoints without embedding system-specific logic into the model itself.

In Protean AI, MCP Servers are first-class, governable resources. An MCP Server exposes executable capabilities, called tools, to AI clients. Each tool defines a clear contract: structured input parameters, deterministic execution logic, and well-defined output.

Protean AI supports two types of MCP Servers.

• Protean for an internal MCP Server
• External for a server hosted outside Protean AI

Internal MCP Servers

Internal MCP Servers are managed and hosted by Protean AI. Protean AI provides a secure, scalable, and reliable runtime for these servers. At the moment, internal servers support only SQL-backed data access.

note

We are working on adding support for MongoDB and other implementations.

External MCP Servers

External MCP Servers are managed outside Protean AI and connected via a network endpoint. They allow you to integrate existing MCP-compatible services without migrating them into the platform.

External servers support:

• URL-based connectivity
• Configurable authentication methods
• Centralized authorization and access control inside Protean

MCP Configuration

Configuring an MCP Server defines how it is identified, connected, secured, and executed within Protean AI. The configuration differs slightly between internal and external servers but follows a consistent structure across both. Every MCP Server requires the following core configuration:

Name

A unique identifier for the MCP Server. This name is used to reference the server across tools, agents, deployments, and authorization policies.

Description

A human-readable explanation of the server's purpose, the systems it connects to, and the types of tools it exposes. This helps users understand when and how the server should be used.

Type

Defines whether the MCP Server is Internal (managed by Protean AI) or External (managed outside Protean and connected via URL).

Timeout

The maximum time Protean AI will wait for a response when executing a tool on this server. This prevents long-running or stalled operations from blocking agent execution.

External MCP Configuration

Apart from the core configuration, external MCP Servers require additional connectivity settings. The following sections describe these settings.

See the screenshot below for an example of MCP Server External configuration.

Snapshot of Protean AI Platform

URL

The endpoint where the external MCP Server is accessible. Protean AI uses this URL to send requests to the server.

Authentication

The authentication method used when communicating with the external server. This defines how the MCP server authenticates requests sent to it from Protean AI. External MCP Servers are not deployed or scaled by Protean AI. Protean acts as a secure gateway and governance layer for access. Protean AI supports the following authentication methods:

• None (unauthenticated requests)
• Token Relay (requests are authenticated using a shared secret)
• API Key (provided in the request headers)
• Basic Auth (username and password)

Internal MCP Configuration

Apart from the core configuration, internal MCP Servers require additional infrastructure and data source settings. The following sections describe these settings.

See the screenshot below for an example of MCP Server Internal configuration.

Snapshot of Protean AI Platform

Datasource Type

In addition to the general configuration, internal MCP Servers require infrastructure and data source settings. Internal MCP Servers must specify a server type. At present, SQL-based servers are supported.

Connection Configuration

Internal MCP Servers of SQL type, connect to their backing database using JDBC. The configuration includes:

• JDBC URL
  The full JDBC connection string, including protocol, host, port, and database name.
• Username
  The database user used by the MCP Server.
• Password
  The password associated with the configured database user.

info

Credentials are stored securely and encrypted. They are not exposed to model or agent code.

Tools

Tools are the executable capabilities exposed by an MCP Server. They define what actions an AI agent can perform.

Each tool consists of:

• A unique name
• A clear, human-readable description
• Input parameters required to execute the tool
• Optional output parameters (external servers)
• An implementation (internal servers)

Tools are discoverable, searchable, and versioned implicitly through the MCP Server they belong to.

Tool Parameters

Parameters define the structured inputs required for a tool to execute successfully.

Each parameter specifies:

• Name and description
• Data type
• Whether it is required or optional

This structure allows agents to call tools reliably while enforcing strict input validation and predictable behavior.

For external MCP Servers, parameters are visible but not editable. For internal servers, parameters are fully configurable.

Tool Implementations

For internal MCP Servers, tools require an implementation that defines how the action is executed. At the moment, only SQL-based tools are supported.

SQL Implementations

Internal MCP Servers support SQL-based tools backed by JDBC connections. Each tool can define a SQL statement that is executed when the tool is called. This enables safe, controlled access to structured data without exposing database credentials or raw connections to AI models.

Deployment and Runtime

Internal MCP Servers are deployable resources. Once configured, they can be deployed to a selected node pool and managed like any other runtime in Protean AI.

Start

Starting an MCP Server runtime initializes the deployment and makes the server available for tool execution. When a runtime is started, the following actions occur:

• Compute resources are allocated
• The MCP Server process is started
• Configured data source connections are initialized
• Tool definitions are loaded and validated
• Health and readiness checks begin

Once started, the MCP Server becomes available for agent tool calls.

Stop

Stopping an MCP Server runtime gracefully shuts down all running instances. When a runtime is stopped, the following actions occur:

• In-flight tool executions are terminated
• MCP Server processes are stopped
• Active connections are closed
• Allocated resources are released

Stopped MCP Server runtimes retain their deployment and configuration settings and can be restarted without recreating the server or its tools. Stopped instances do not consume compute resources and do not restart automatically. However, if a running instance becomes unhealthy and terminates unexpectedly, it will be restarted automatically while the runtime is in a running state.

Scale

Scaling controls how many MCP Server runtime instances are active for a deployment. Scaling can be adjusted at any time while the runtime is running.

Scale up

Scaling up increases the number of active MCP Server instances. Use scale up when:

• Tool execution volume increases
• Concurrent agent calls grow
• Lower execution latency is required
• Higher availability is needed

Scale down

Scaling down reduces the number of active MCP Server instances. Use scale down when:

• Tool usage decreases
• Reducing operational cost is a priority
• Lower resource utilization is sufficient

Scaling operations are applied dynamically and do not require changes to server configuration, tools, or connected data sources.

Delete

Deleting an MCP Server runtime permanently removes the deployment. When a runtime is deleted, all running MCP Server instances are stopped and the deployment configuration is removed. The MCP Server definition, including its tools, parameters, and authorization settings, are also deleted.

tip

To delete a deployment, and not underlying MCP Server definition, use stop or scale down instead.

MCP Server Runtime Deletion

• An MCP Server runtime can only be deleted if it is not currently being used by any Agent.

Caution

Deletion is irreversible. MCP Server runtime instances and deployment configuration cannot be recovered.

Logs and Events

Logs and events provide operational visibility into runtimes. They help you monitor execution, diagnose failures, and understand how a deployment behaves over time at both instance and deployment scope.

Logs

Runtime provides detailed observability to help diagnose issues and understand runtime behavior. Logs can be accessed at instance levels, individual runtime instances. Logs include inference errors and warnings

Events

Events provide a structured view of significant runtime actions and state transitions. Events are organized into two categories:

• Instance-level events show lifecycle and execution events for a specific instance
• Deployment-level events summarize changes affecting the entire runtime

Events include:

• Scaling operations
• Scheduling decisions
• Health and readiness state changes

Access Control

Access Control in Protean AI governs who can view, create, modify, and operate resources across the platform. It is designed for enterprise environments where security, isolation, and governance are mandatory.

Protean AI follows a principle of least privilege, ensuring users and systems are granted only the permissions required to perform their tasks.

Role→ Action↓	Admin	Model Admin	User	Owner	Viewer	Description
Create	Yes	No	No	NA	NA	Register / Create an MCP server
Read	Yes	No	No	Yes	Yes	View MCP servrer and use it from agents
Update	Yes	No	No	Yes	No	Update MCP server metadata, add and remove tools
Delete	Yes	No	No	Yes	No	Remove MCP server, its deployment and definition
Manage Access	Yes	No	No	Yes	No	Grant or revoke permissions for users and groups.

Workflow

1. Add MCP Server and Choose the MCP Server type (Internal or External).
2. Configure the server settings, including description, timeout, and connection details.
3. (Internal only) Configure the backing data source and deployment settings, such as node pool selection.
4. Define or review the tools exposed by the MCP Server, including parameters and implementations.
5. Review the configuration and authorization settings.
6. Save the configuration to create the MCP Server deployment.
7. This creates a deployment resource in the system with 0 desired instances.
8. Scale up the deployment to start MCP Server runtime instances and make tools available for execution.
9. Use the MCP Server from AI agents.

Result

After the MCP Server deployment is created, it can be started as a runtime. When started, the server is deployed onto the selected nodes and becomes available for tool execution by agents. With start, stop, scaling, and deletion controls, MCP Server Runtime provides operational flexibility while maintaining clear governance and visibility. Combined with runtime status, instance-level logs, and execution events, MCP Servers can be operated confidently in production, ensuring reliable tool execution,