MCP-Compatible AI Platform 'mitoco Buddy' Upgraded

Terrasky Inc. (Headquarters: Chuo-ku, Tokyo; Representative Director, CEO, and President: Hidetaka Sato, hereinafter 'Terrasky') will implement an upgrade to its MCP (Model Context Protocol)-compatible AI platform 'mitoco Buddy' on Tuesday, July 14, 2026.

The latest version includes numerous enhancements and improvements.

New features have been added to make 'mitoco Buddy' more convenient and powerful, including 'Auto Router' for selecting the optimal model based on request content, the ability to import files from SharePoint into knowledge bundles, and a workflow function requiring approval before publishing created skills.

Additional features include dynamically generating skills within chat, searching past chat history, and resuming conversations from previous interactions.

The platform also now supports the latest model 'Claude Sonnet 5'.

Main upgrade contents are as follows:

Auto Router Function

Auto Router analyzes the complexity and cost of user requests to automatically select the most suitable AI model. This eliminates the need for users to manually switch between models, ensuring optimal performance for each use case.

Based on the request content, one of the following three levels is automatically selected:

・light (e.g., claude-4.5-haiku): Suitable for short, independent requests that can be resolved in a single interaction.

・balanced (e.g., claude-4.6-sonnet): Designed for business communication where users can use the output with minimal editing. Also suitable for consecutive short conversations or interpreting results from read-only tools.

・heavy (e.g., gpt-5.4): Used for complex code or data analysis, long conversations requiring context from previous exchanges, or tasks with high opportunity cost upon failure. Also ideal when a large number of tool calls are required.

The model selection logic will continue to be improved. Notably, the actual content of user conversations is never referenced during this selection process.

SharePoint File Import to Knowledge Bundles

'Knowledge Bundles,' structured data optimized for AI understanding, now support direct file import from SharePoint. This eliminates the need to manually download and upload documents from SharePoint, enabling seamless synchronization with agents through simple operations.

Agent Publishing Approval Workflow

A new workflow function requires pre-approval by administrators or owners before agents created within an organization are shared company-wide. This introduces a third-party review process before publication, enhancing agent quality, security, and governance.

Skill Suggestion Based on Past Conversations

The AI detects repetitive tasks or complex interactions conducted by users in general chat and suggests them as reusable 'skills' for efficiency.

This eliminates the need to repeatedly type the same questions or instructions, enabling complex operations to be executed with a single click.

Dynamic Skill Generation within Chat

This feature allows necessary skills to be automatically generated on the spot through conversation in general chat, eliminating the need to upload files.

Users can seamlessly prepare new skills without leaving the chat interface and immediately execute them.

Dynamic MCP Tool Configuration and Auto-Loading

Based on chat content and user requests, the AI can search for required MCP tools, suggest adding them, or automatically load and execute them.

Chat History Search Function

Past chat histories can now be searched using natural language queries such as 'What conversations did I have this week?'

The AI searches relevant conversations from the past three months (approximately 93 days), summarizes the content, and allows users to reopen past chats from the results to immediately resume conversations with previous context intact.

Provision of MCP Connection URLs

Dedicated connection URLs (web addresses) are issued for each type of MCP (Model Context Protocol) used to integrate with external services or internal systems.

By simply pasting the provided connection URL into the external system, secure and reliable integration can be established.

OAuth Token Management

Users can now securely manage and optimize authentication credentials (OAuth tokens) for integration with external applications. This allows elimination of unnecessary access rights, strengthening security.

Addition of New MCP Integration Partners

Integration is now supported with the following services. Everyday business applications can now connect with mitoco Buddy, enabling direct operation from chat or agents.

Azure Files

From general chat, personal/organizational agents, or automation workflows, users can directly access 'Azure Files,' the company's file-sharing service, to search and retrieve data according to their permissions on behalf of the user. This significantly reduces the effort required to locate needed information.

IBM Netezza

Leveraging IBM Netezza's capability for high-speed analysis and processing of large datasets, users can now seamlessly access large internal databases directly from the regular chat interface. Without switching to dedicated tools, rapid information extraction from massive datasets and automation of operations incorporating data integration can be easily achieved.

Support for Latest AI Models

To enable more advanced reasoning and natural conversations, support has been added for the latest AI models.

・Claude Sonnet 5

For detailed information on the above features, please refer to the release notes.

https://www.terrasky.co.jp/files/mitoco_Buddy_ReleaseNote_Ver.20260714.pdf

＜About mitoco Buddy＞

'mitoco Buddy' is an MCP-compatible AI platform that supports enterprise data utilization and business automation. It can integrate with approximately 50 services, including Salesforce and Slack. Users can flexibly create their own agents, enabling business automation within existing multi-cloud environments.

*Company names and service names mentioned in this text are registered trademarks of their respective owners.