feat: add --stdio-env option for passing environment variables

[ddnowicki]

Add support for passing environment variables to the upstream stdio subprocess. This enables use cases where the upstream MCP server requires environment variables (e.g., API tokens, URLs) that need to be passed through the filter.

Changes:

• Add stdio_env field to UpstreamConfig model
• Add --stdio-env KEY=VALUE CLI option (repeatable)
• Add MF_STDIO_ENV environment variable support (semicolon-separated)
• Pass env vars to NpxStdioTransport, PythonStdioTransport, and StdioTransport

Usage:
mcp-filter run -t stdio --stdio-command npx \ --stdio-arg "-y @some/mcp-server" \ --stdio-env "API_KEY=secret" \ --stdio-env "API_URL=https://example.com"

[ddnowicki]

Hi @pro-vi,

I was trying to use mcp-filter with Claude Code to proxy a GitLab MCP server (@zereight/mcp-gitlab) that requires environment variables like GITLAB_PERSONAL_ACCESS_TOKEN and GITLAB_API_URL.

The problem: The env section only sets environment variables for mcp-filter itself, but these variables are not forwarded to the spawned subprocess (npx @zereight/mcp-gitlab).

The upstream GitLab MCP server failed with:
GITLAB_PERSONAL_ACCESS_TOKEN environment variable is not set

Solution

I added a --stdio-env option that allows passing environment variables directly to the upstream subprocess:

mcp-filter run -t stdio --stdio-command npx
--stdio-arg "-y @zereight/mcp-gitlab"
--stdio-env "GITLAB_PERSONAL_ACCESS_TOKEN=glpat-xxx"
--stdio-env "GITLAB_API_URL=https://gitlab.example.com/api/v4"
-a "get_merge_request,mr_discussions"

The implementation uses the existing env_vars/env parameters that fastmcp transports already support (NpxStdioTransport, PythonStdioTransport, StdioTransport).

Changes

• config.py: Added stdio_env: Dict[str, str] field + MF_STDIO_ENV env var parsing
• cli.py: Added --stdio-env KEY=VALUE CLI option (repeatable)
• upstream.py: Pass env to transport constructors

Could you please review this PR? This feature would enable mcp-filter to work with any MCP server that requires environment variables.