Skip to content

mcp-tools.md

Latest commit

 

History

History
122 lines (83 loc) · 4.13 KB

mcp-tools.md

File metadata and controls

122 lines (83 loc) · 4.13 KB

MCP Tools

LinkWork provides AI workers with external capabilities through the MCP Tool Bus. Compatible with the Model Context Protocol (MCP) standard, all external tools are accessed through a unified proxy layer that automatically handles auth, routing, and usage metering.

Design Philosophy

To accomplish real work, AI workers need more than just LLM reasoning — they need to query databases, access APIs, operate filesystems, control browsers, and more.

LinkWork chose MCP as the tool integration standard:

  • Protocol Standardization — Follows the MCP open standard with good ecosystem compatibility
  • Unified Proxy — All tool calls go through the gateway proxy, no direct connections to external services
  • Auth Governance — Tool-level permission control, admins decide who can use which tools
  • Observable Usage — Every tool call is recorded, supporting metering and auditing

Three-tier Tool Architecture

linkwork-mcp-gateway (Tool Gateway)
  ├── Auth & Routing
  ├── Health Checks
  └── Usage Metering
        │
        ▼
  MCP Tool Services
  ├── Database Query Tool
  ├── API Call Tool
  ├── File Operation Tool
  ├── Browser Control Tool
  └── ...more tools

linkwork-mcp-gateway

The MCP tool gateway is the sole entry point for all tool calls, responsible for:

Function Description
Tool Registration & Discovery Manage all registered MCP tool services
Auth Proxy Control tool access by role permissions
Health Checks Periodic probing, auto-offline unhealthy tool services
Usage Metering Record call duration, results, and costs for each tool invocation
Rate Limiting Per-service QPS limits to prevent abuse

Tool Types

LinkWork currently supports the following types of MCP tool services:

Type Description Examples
HTTP Remote HTTP API services Jira, GitLab, database queries
SSE Remote Server-Sent Events services Real-time monitoring, streaming data sources

More types (e.g., stdio local processes) will be supported in future releases.

Tool Lifecycle

Registration

Admins register MCP tool services through the management interface, providing:

  • Service name and description
  • Service type (HTTP / SSE)
  • Service address
  • Auth configuration (API Key, Token, etc.)
  • Health check endpoint (optional)

Health Checks

After registration, the platform automatically performs periodic health checks:

Status Meaning
Online Recent probe successful, response normal
Degraded Response latency high or intermittent failures
Offline Multiple consecutive probe failures

Role Association

Assign tools to specific roles:

  • A role can be associated with multiple MCP tools
  • Tool configuration is baked into the image during role building
  • AI workers can only use authorized tools at runtime

MCP Factory

The MCP Factory is part of the AI supply chain, providing unified tool management and governance:

  • Tool Registration & Discovery — Unified tool catalog, searchable by category and tags
  • Version Management — Tool service version tracking
  • Health Monitoring — Real-time awareness of all tool service availability
  • Usage Analytics — Call volume, latency, error rate, and other statistics

Security Considerations

MCP tool integration follows LinkWork's security principles:

  • Least Privilege — Each role can only access authorized tools
  • Credential Isolation — Tool auth credentials are managed by the platform, invisible to AI workers
  • Network Governance — Tool calls go through the gateway proxy, AI workers cannot directly connect to external services
  • Audit Trail — Every tool call has a complete audit record

Further Reading