Connectors Overview

Connect your agents and workflows to external services through MCP servers.

What Are Connectors?

Connectors are MCP (Model Context Protocol) servers that expose tools from external services — Linear, Slack, Google, GitHub, and more. inference.sh manages authentication and tool execution so your agents can use these services directly.

1Your Agent2  ├── app tool (image generation)3  ├── call tool (REST API)4  └── connector tool (Linear: create issue)5        ↓6    inference.sh handles auth + MCP protocol7        ↓8    Linear MCP Server

Three Ways to Use Connectors

1. CLI — Call Tools Directly

Browse servers, connect, and call tools from your terminal:

1belt mcp list2belt mcp connect linear3belt mcp run linear:list_issues --input '{"project": "INF"}'

2. Agent Tools — Give Agents Access

Add connector tools to your agents. The agent decides when to use them:

1tools:2  - name: create_issue3    type: mcp4    description: Create a Linear issue5    mcp:6      integration_id: "integ_abc123"7      tool_name: "create_issue"

3. MCP Server — Connect External Clients

Connect Claude Code, Cursor, or any MCP client to inference.sh. Your connected services are available as proxy tools:

1{2  "mcpServers": {3    "inference": {4      "url": "https://api.inference.sh/mcp",5      "headers": { "Authorization": "Bearer inf_your_key" }6    }7  }8}

How Authentication Works

Each connector server declares its auth type:

Credentials are encrypted at rest and injected at runtime. Your agents never see raw tokens.

Categories

Next

• Browsing & Connecting — discover and set up connectors
• Using Tools — call tools from the CLI
• Agent Tools — add connectors to your agents
• MCP Server — use inference.sh as an MCP server