Tokeny comes with built-in core tools like file read/write, Shell commands, and content search -- these are enough for a wide range of daily tasks. But the world of tools extends far beyond these -- you might need the AI to open a browser and fill out forms, query a database, or call a SaaS platform's API. This is why the MCP tool protocol exists: to make the AI's toolbox infinitely extensible.
If Skills are the AI's "professional knowledge," then MCP tools are the AI's "hands" -- through the MCP protocol, the AI can interact with various external services and tools, truly transforming from "can only talk" to "can take action."
What Is MCP
MCP stands for Model Context Protocol, an open standard that defines how AI models discover and invoke external tools. Think of it as "USB for AI" -- any tool service that follows the MCP standard can be "plugged into" the AI, giving it new capabilities.
This protocol was initiated and open-sourced by Anthropic and has gained broad community support. As of early 2026, community-contributed MCP services cover search engines, browser automation, database operations, file format processing, code repository management, messaging notifications, and many other areas. The ecosystem is growing rapidly.
How MCP Works
MCP uses a client-server architecture, with a straightforward overall flow:
- 1Tokeny (the client) connects to an MCP service and retrieves the list of tools and their descriptions provided by that service
- 2During a conversation, the AI identifies the need for an MCP tool (e.g., web search) and automatically initiates a tool call
- 3The MCP service receives the call request, performs the specific operation (e.g., actually querying a search engine), and returns the results
- 4The AI receives the tool's results and continues with subsequent analysis or conversation
The entire process is transparent to the user -- you can see which tools the AI called, what parameters it passed, and what results were returned in the conversation. There are no invisible operations.
Two Connection Types
MCP services support two connection methods, suited for different scenarios:
stdio (Standard Input/Output) -- the MCP service runs as a local process on your computer. Tokeny communicates with the process through standard input/output. This is the most common type; most community MCP services (such as Playwright and file processing tools) use this approach. Communication uses a local process, but the service can still access the network. Data boundaries depend on its implementation and configuration.
SSE (Server-Sent Events) -- the MCP service runs on a remote server, and Tokeny connects via HTTP. This is suited for scenarios requiring remote computing resources, such as GPU services running in the cloud or internal tools shared by a team.
For most individual users, stdio-type MCP services are the preferred choice -- using local-process communication, with network behavior still requiring review. SSE is better suited for team collaboration or advanced scenarios requiring remote resources.
Tokeny's Built-in MCP Services
Tokeny comes with two MCP services built in and ready to use, requiring no additional configuration:
- WebFetch -- retrieves the content of any web page and converts it to structured text. The AI can "read" any publicly accessible page on the internet and extract key information. Includes built-in SSRF protection to block requests to internal network addresses
- Web Search -- searches for information through search engines and retrieves the latest web content. This is the foundational capability for skills like Deep Research and Web Search
These two built-in services cover the most essential online needs. For users who don't need complex tool integrations, this is already sufficient.
Adding Third-Party MCP Services
Let's walk through adding the Playwright MCP (browser automation) as an example:
- 1Open the "Settings" panel at the bottom-left of the app, navigate to the "MCP Services" tab
- 2Click the "Add Service" button
- 3Select connection type stdio (Playwright is a local service)
- 4Enter the startup command in the command field: npx @playwright/mcp@latest
- 5Save the configuration -- the system automatically attempts to connect
Once connected, you'll see the service status change to "Connected" on the MCP management page, along with all the tools it provides. From then on, the AI can use these tools in conversations -- for example, if you say "Help me open Google and search for today's weather," the AI will call Playwright's browser tools to complete the task.
For SSE-type remote services, the process is the same, except in step 4 you enter the service URL (e.g., https://mcp.example.com/sse).
Tool Deduplication and Priority
When using multiple MCP services simultaneously, tool name conflicts may occur -- for example, two services both providing a tool named "search." Tokeny has clear rules for handling this:
- Built-in tools take priority -- Tokeny's built-in tools (file read/write, Shell, search, etc.) always take precedence over MCP tools with the same name
- First registered, first served -- when multiple MCP services provide tools with the same name, the first-connected service takes priority
- Automatic deduplication -- duplicate tool names are automatically filtered to prevent confusion for the AI
This means adding new MCP services won't affect the normal operation of existing tools. You can safely try various MCP services without worrying about tool conflicts.
Auto-Reconnection and Stability
MCP services may disconnect during use due to network fluctuations or process exceptions. Tokeny has a built-in auto-reconnection mechanism:
- Upon detecting a disconnection, the system automatically attempts to reconnect
- Uses exponential backoff strategy (gradually increasing intervals) to avoid resource-intensive frequent reconnections
- After successful reconnection, tool availability is automatically restored with no manual intervention needed
- If reconnection fails multiple times, error status and detailed logs are displayed on the MCP management page
Environment Detection: One-Click Dependency Check
stdio-type MCP services (like Playwright) depend on the local runtime environment -- typically requiring Node.js, and sometimes Python or Git. If you're unsure whether your computer meets the requirements, you can use Tokeny's built-in environment detection feature.
On the MCP management page, click the "Environment Detection" tab. The system will automatically check the installation status and version information of key dependencies like Bash/Shell, Node.js, Python, and Git. The results will clearly show what's installed, what's missing, and recommended installation methods.
Security Considerations
MCP services are essentially programs running on your computer (or connected to a remote server). They have certain system permissions -- they can access the network, read/write files, and execute commands. Therefore, security awareness is important:
- Only install MCP services from trusted sources (official GitHub repositories, well-known community projects, etc.)
- Review the project's README and source code before installation to understand its functionality and permission scope
- For SSE-type remote services, ensure the connection URL points to a server you trust
- Periodically review your list of installed MCP services and remove those no longer in use
The security of third-party MCP services is the responsibility of the service provider -- Tokeny cannot guarantee their behavior. Before installing unfamiliar MCP services, we recommend reviewing their source code and community feedback on GitHub.
Remember: Tokeny's built-in tools (file read/write, Shell commands, content search, etc.) do not depend on MCP and work out of the box. MCP provides additional extensibility on top of built-in capabilities. Even without adding any MCP services, Tokeny's core functionality is fully available.