Flyte MCP: native MCP to access flyte remote commands and more.

[cosmicBboy]

This pull request introduces support for deploying and serving Model Context Protocol (MCP) servers within Flyte, enabling AI assistants and LLM-based clients to interact with Flyte via standardized MCP tools. It adds new example applications, a generic MCP app environment, a CLI entrypoint, and corresponding tests. The main themes are: new Flyte MCP app environments, CLI and packaging updates, and test coverage.

New Flyte MCP App Environments and Examples:

• Added FlyteMCPAppEnvironment and MCPAppEnvironment classes to allow serving Flyte-integrated MCP servers and generic FastMCP servers, respectively. These environments expose Flyte operations as MCP tools over HTTP with configurable endpoints, tool groups, and search paths. (src/flyte/ai/mcp/_mcp_app.py [1] src/flyte/ai/mcp/__init__.py [2]
• Added comprehensive examples for deploying MCP servers:
  • flyte_mcp_app.py: Deploys a full-featured MCP server with all Flyte tools.
  • flyte_mcp_app_filtered.py: Shows how to restrict tool access and configure search paths with an allowlist.
  • generic_mcp_app.py: Demonstrates serving a generic FastMCP server as a Flyte app.

CLI and Packaging Updates:

• Added a new CLI entrypoint flyte-mcp for serving a Flyte MCP server, with options for tool selection, search paths, and HTTP configuration. (src/flyte/_bin/mcp.py [1] pyproject.toml [2]
• Updated pyproject.toml to include a new mcp optional dependency group for examples and to register the new CLI. [1] [2]

Test Coverage:

• Introduced unit tests for the new MCPAppEnvironment, verifying correct mounting, health checks, and error handling for invalid configurations. (tests/flyte/ai/mcp/test_mcp_app_environment.py tests/flyte/ai/mcp/test_mcp_app_environment.pyR1-R63)

[kumare3]

i feel this is dangerous. but i guess you will not import it till we import flyte.ai.mcp right