MCP Native Support

The Model Context Protocol (MCP) is emerging as a standard for tool execution. But running MCP servers directly means managing server instances, handling auth separately, and losing centralized audit.

Hexel Tool Gateway supports MCP natively as an execution backend type. You get the MCP ecosystem with the gateway's reliability and governance on top.

How it works:

Register an MCP server as a backend: - The gateway manages server instances (mcp_server_instances table) - Tool calls are routed to the MCP server using the standard protocol - Credentials, rate limiting, dedup, and audit all work identically

Same interface for your agents:

# Your agent doesn't know if this is HTTP or MCP
tools = client.get(toolkits=["MY_MCP_TOOLKIT"])
results = client.handle_tool_calls(openai_response)

Whether the backend is a REST API, a GraphQL endpoint, or an MCP server — your agent code is the same single line. The gateway handles the protocol translation.

Why this matters:

• MCP tools get the same OAuth, rate limiting, dedup, and audit as HTTP tools
• You can mix HTTP and MCP backends in the same toolkit
• Migration between protocols requires zero agent code changes
• New MCP tools from the community work instantly through your gateway

Before: Run MCP servers directly. No auth management. No audit. No rate limiting. Every server is a standalone process to manage.

After: Register the MCP server once. Everything else — credentials, limits, audit, dedup — is handled by the gateway.