Introduction to MCP
The Model Context Protocol (MCP) is an open standard designed to enable seamless, secure communication between LLM clients and external data sources or tools. Datafuse acts as an active, high-availability bridge for MCP ecosystems.
Why MCP Matters for AI Agents
Traditionally, every AI model or agent framework has defined its own custom way of loading context, formatting prompts, and executing tool actions. This fragmentation forces developers to write unique, fragile bindings for each new model or framework they want to support.
MCP solves this by introducing a standardized contract:
- Context Standardization: Normalizes how data, files, and live context are formatted and passed to the model.
- Decoupled Servers: Separates the data provider logic from the core LLM execution block.
- Universal Tooling: Any MCP-enabled client (such as Claude Code, cursor, or custom LLM runners) can immediately invoke any integration compiled by Datafuse.
Single Toolkit MCP vs. Direct Mode
Datafuse supports two flexible integration topologies for Model Context Protocol systems:
1. Direct Integration Mode
In this configuration, your application uses the standard Datafuse SDK to select and execute tools individually. You have fine-grained programmatic control over execution variables, custom exception handling, and model choices.
2. Single Toolkit MCP Mode (Bridge Mode)
Datafuse compiles an entire integration provider (e.g. GitHub or Jira) directly into a standalone MCP Server. Your application or workspace client connects to Datafuse via standard MCP transport rules.
Datafuse then exposes all associated actions as native MCP tools, managing user credentials behind the scenes.
| Feature Comparison | Direct Integration Mode | Single Toolkit MCP Mode |
|---|---|---|
| Protocol | Datafuse REST / SDK | Standard MCP Protocol (JSON-RPC) |
| Client Support | Custom App, LangChain, CrewAI | Cursor, Claude Desktop, Claude Code |
| Credential Handling | Vault-decoupled in SDK | Completely hidden at bridge boundary |
| Tool Grouping | Granular execution | Unified server toolkit (bundled tools) |
Seamless Model Context Ingress
By bridging MCP systems, Datafuse allows models to pull live documentation, browse files, and query local context seamlessly during agent executions:
!NOTE When you register an MCP server inside Datafuse, all prompt templates, custom resource schemas, and action parameters are instantly translated to our platform OpenAPI format, making them usable in non-MCP standard environments as well.