MCP Servers

Connect agents to external tools and data sources via the Model Context Protocol.

What Is MCP?

The Model Context Protocol (MCP) is an open standard for connecting AI agents to external tools and data sources. Think of it as a universal adapter — instead of building custom integrations for every external system an agent might need to access, MCP provides a single, standardized way to make any tool or data source available to any agent.

When you register an MCP server with the platform, the tools it exposes become available for assignment to any agent. The agent does not need to know the details of how the external system works — it interacts with the MCP server through a consistent interface, and the server handles the translation to the underlying system.

How MCP Servers Become Agent Tools

The process for connecting an external tool via MCP is straightforward:

Register the MCP server. In the admin UI, create a new MCP server registration. Provide the server's connection details — its address, transport type, and any authentication credentials it requires.
The platform discovers available tools. Once connected, the platform queries the MCP server to learn what tools it offers. Each tool has a name, description, and input schema that describes what parameters it accepts.
Assign tools to agents. The MCP server's tools appear in the tool selection interface when configuring agents. Assign any combination of MCP tools to any agent, just like you would assign built-in tools.
Agents use the tools. During a conversation, the agent can call any of its assigned MCP tools. The platform routes the request to the appropriate MCP server, which executes the operation and returns the result.

MCP servers are registered once and can provide tools to many agents. If you register a server that exposes 10 tools, any agent in the system can use any combination of those tools.

Transport Types

MCP supports multiple transport mechanisms for communicating between the platform and external servers. The right choice depends on how the MCP server is deployed and what kind of connectivity is available:

Transport	How It Works	Best For
SSE (Server-Sent Events)	The platform connects to the MCP server over HTTP using Server-Sent Events for real-time communication. The server must be accessible over the network.	Remote MCP servers, cloud-hosted tools, servers shared across multiple environments.
STDIO	The platform launches the MCP server as a local process and communicates via standard input/output streams. The server runs on the same machine as the platform.	Local tools, development and testing, tools that need direct filesystem access.
Streamable HTTP	A newer HTTP-based transport that supports bidirectional streaming. Combines the network accessibility of SSE with improved efficiency and reliability.	Production deployments where you need reliable, high-performance communication with remote MCP servers.

Built-In Bosca MCP Server

The platform includes a built-in MCP server that exposes the Bosca GraphQL API as AI-friendly tools. This means agents can interact with the platform's own data and capabilities through the same MCP interface used for external tools.

The built-in MCP server provides several tools:

List fields — Discover what fields are available on any GraphQL type, helping the agent understand what data it can access.
Introspect types — Examine the structure of GraphQL types, including their fields, arguments, and relationships.
Search schema — Find relevant types and fields by searching the GraphQL schema with natural language descriptions.
Execute queries — Run GraphQL queries against the platform's API to retrieve data, making it possible for agents to access content, analytics, user information, and any other data exposed through the API.

This built-in server is particularly useful for building agents that work with your platform data. A "Content Insights" agent could use these tools to query content metadata, analyze publication patterns, and generate reports — all through natural conversation.

The built-in Bosca MCP server is a great way to let agents access your platform data without building custom tools. Since it works through the standard GraphQL API, agents benefit from the same permission system and access controls that apply to all API access.

CLI MCP Integration

The Bosca CLI includes MCP support that exposes Work Ops tools for AI coding assistants. This means AI tools like coding assistants and IDE integrations can manage your operational workflow directly through the MCP protocol.

Through the CLI MCP integration, AI assistants can:

Manage portfolios — Create, update, and organize portfolios that group related projects together.
Work with projects — Create projects, update their details, and track their status.
Handle tasks — Create, assign, update, and complete tasks within projects.
Manage sprints — Plan and track sprints, move tasks between sprints, and review sprint progress.
Organize boards — Work with kanban-style boards to visualize and manage work in progress.

This integration bridges the gap between your AI development tools and your project management workflow. A coding assistant can create a task for a bug it discovers, update a task's status when it completes a fix, or check what tasks are assigned to the current sprint — all without leaving the development environment.

Managing MCP Servers

MCP server registrations are managed through the admin UI. The lifecycle is simple:

Create a registration. Provide the server name, transport type, connection details, and any required credentials. The platform tests the connection and discovers available tools.
Enable or disable. You can disable an MCP server without deleting its registration. Disabling a server makes its tools unavailable to agents but preserves the configuration so you can re-enable it later. This is useful for maintenance windows or when troubleshooting issues.
Update connection details. If a server's address, credentials, or other connection parameters change, update the registration in the UI. Agents will use the new connection details on their next request.
Remove a registration. Delete a registration when you no longer need the server. Any agents that had the server's tools assigned will lose access to those tools.

Before disabling or removing an MCP server, check which agents use its tools. Removing a server's tools from an active agent may cause the agent to fail or produce incomplete responses when it tries to use a tool that is no longer available.

Common Use Cases

MCP servers open up a wide range of integration possibilities. Here are some common patterns:

Use Case	How It Works
External databases	Register an MCP server that connects to an external database (PostgreSQL, MySQL, BigQuery, etc.). Agents can query the database using natural language, and the MCP server translates requests into SQL and returns results.
Third-party APIs	Wrap any REST or GraphQL API as an MCP server. Agents can then interact with services like CRM systems, payment platforms, shipping providers, or social media APIs through natural conversation.
Knowledge bases	Connect to vector databases, document stores, or search engines. Agents can retrieve relevant information from your organization's knowledge base to ground their responses in real data.
Custom tooling	Build MCP servers for your organization's internal tools — deployment systems, monitoring dashboards, content pipelines, or any custom workflow. This lets agents interact with tools that are unique to your organization.
Development workflow	Use the CLI MCP integration to connect AI coding assistants to your project management system. Manage tasks, sprints, and projects without switching contexts.

The MCP ecosystem is growing rapidly. Many popular tools and services now offer MCP-compatible servers. Before building a custom integration, check if an existing MCP server already provides the capability you need.