MCP FOR AI AGENTS
AutronAI uses Model Context Protocol to bring external tools and services into the Agent system. Explore available MCP servers, install your own connection, configure the required authentication, inspect its tools, and give a Tool Agent access to the specific capabilities required for its responsibility.
Beyond the model
A model can understand and generate text, but many tasks require access to external systems. MCP supplies a consistent connection layer; it is not the Agent itself.
Bring an external service into the workspace through an MCP server definition and an installed instance.
Review its authentication requirements and available capabilities.
Give a Tool Agent only the tools required for its responsibility.
Connection architecture
The server and installed instance are the connection and capability source. A Tool Agent receives selected tools. A Chatbot is the higher-level composition that brings focused Agents together.
The provider or system where the capability lives.
The reusable definition for connecting to that capability source.
The configured workspace connection created from a server template.
The individual actions chosen from the instance catalog.
The focused Agent that receives those selected capabilities.
The conversational composition that can bring multiple Agents together.
Explore
The current marketplace separates discovery from workspace connections with Explore and Installed tabs. Search, filter by category, sort the catalog, and compare tool counts and credential requirements before opening a server.
Use search, category filters, and sorting to narrow the catalog.
Cards expose tool count and indicate when credentials are required.
Move between servers to explore and connections already installed.
Preview
The preview shows the server description and slug, authentication scheme and categories, plus the name and description of each exposed tool.
Connection instance
Give the instance a recognizable name and provide the configuration required by its server template. The installed instance is the capability source that later appears for Tool Agents.
A reusable server template with connection requirements and available tool definitions.
A named, per-workspace configured connection created from that template.
Authentication
The current model supports none, api_key, and oauth. API-key servers expose required credential fields and can render the key into a bearer header when required. OAuth is server-specific: the marketplace currently has dedicated installation and reconnection handling for Google Workspace OAuth, rather than assuming every server supports OAuth.
No credential input is required by the server definition.
Enter the required API key through its dedicated credential field.
Complete provider authorization only where the server supports it.
Connection data
Installation sends ordinary settings as config_values and sensitive credential fields as secret_values. Secret fields render as password inputs, separate from normal configuration.
Installed workspace
The Installed view shows each instance name, source server, server tool count, and whether credentials are configured, with controls to inspect or remove the connection.
Explicit assignment
Installation makes an instance available in the workspace. Tool access is granted later while creating or editing a Tool Agent.
Create the configured MCP instance.
The instance becomes a capability source.
Define the focused responsibility.
Choose the relevant connection.
Grant only the required actions.
The Agent receives the selection, not the entire catalog.
Tool selection
The selector groups selected and unselected tools, shows descriptions and a selected count, and supports individual selection, select all, and clear.
Granular actions
A server catalog is not an indivisible permission set. Tools remain individually assignable, so a Tool Agent can receive a smaller subset that matches its job.
Clear concepts
Keeping responsibility, procedure, connection, and action separate makes the system easier to understand.
Agent Instructions
Define its role, boundaries, and expected behavior.
Skill
Provides reusable procedural guidance.
MCP Instance
Represents the configured workspace connection.
MCP Tool
Represents one capability from that system.
Procedure plus action
Tool Agents can also have Skills. Instructions define responsibility, a Skill supplies reusable procedure, and an MCP tool supplies an external action. A Skill is guidance, not executable external code.
Tool Agent
Responsibility and operating boundaries.
Tool Agent
A reusable procedure selected for the task.
Tool Agent
An external action the Agent may invoke.
System composition
MCP tools are one part of AutronAI. Focused Agents can work with Structured Data, Documents, Skills, Memory, or tools. Chatbots compose those capabilities, and Channels connect the experience to supported communication surfaces.
Right capability, right Agent
Not every Agent needs MCP. A Document Agent may use documents, a Structured Data Agent may use a structured source, and a Custom Agent may use Skills. A Tool Agent is designed to receive selected external MCP tools.
Intentional access
Choose the smallest useful tool set.
Define the Tool Agent's job before assigning capabilities.
Keep unrelated capabilities in different focused Agents.
Inspect each tool before adding it.
External dependencies
Provider availability, permissions, account state, API or provider limits, credential validity, and provider policy changes can affect what a connected tool can do.
Custom definitions
The customer MCP workspace supports owner-scoped external server definitions with a name, slug, description, URL template, headers template, authentication scheme, enabled state, and—when relevant—a credential help URL or OAuth provider and scopes. Current customer-created definitions use owner-only visibility.
Catalog maintenance
The workspace includes an explicit synchronization flow. It previews the tools a server exposes, shows added, updated, and removed counts, and lets an operator review before approving the catalog update. Synchronization is deliberate, not continuous or automatic.
End-to-end workflow
Find an MCP server.
Review tools and authentication.
Create a named instance.
Provide credentials or complete OAuth where supported.
Define its responsibility.
Give it only the required tools.
Attach the Tool Agent to a Chatbot.
Controlled capability
MCP makes external capabilities easier to connect, but the goal is not to maximize the number of available tools.
The goal is to give each Agent a controlled set of capabilities that matches its responsibility.
A connection can expose many tools. A Tool Agent can use a smaller subset. Skills can describe reusable procedures. Other Agents can handle documents or structured data. The Chatbot can then compose the capabilities that belong together.
That structure keeps the AI system understandable as it grows.
FAQ
Questions, answered
Model Context Protocol provides a consistent way to expose external capabilities to AI systems. In AutronAI, a workspace installs or configures an MCP instance, reviews its tools, and assigns selected actions to a focused Tool Agent.
No. Installation makes the connection available in the workspace. Tool access is granted later by selecting an MCP instance and individual tools for a Tool Agent, so an installed catalog does not become a broad permission set automatically.
The current model supports none, api_key, and oauth authentication schemes. OAuth behavior is server-specific; the present marketplace includes dedicated Google Workspace OAuth handling rather than assuming a universal OAuth flow for every server.
An MCP tool represents an external action that an Agent may invoke. A Skill represents reusable procedural guidance. Instructions define the Agent's responsibility, while Skills and tools contribute different parts of how that responsibility is fulfilled.
The customer MCP workspace supports owner-scoped external server definitions with connection, authentication, and tool-catalog configuration. Catalog refresh is an explicit preview-and-approval workflow rather than a background process.
Build with AutronAI
Install an MCP connection, inspect the tools it exposes, and give each Tool Agent the capabilities required for its job.
