CLI usage¶

The CLI can serve a compressed MCP server or start helper modes that create local command/code clients.

Standard MCP proxy¶

mcp-compressor -c medium -- python server.py

Compression level shorthand:

mcp-compressor -c low -- python server.py
mcp-compressor -c medium -- python server.py
mcp-compressor -c high -- python server.py
mcp-compressor -c max -- python server.py

Custom server name¶

mcp-compressor -c medium --server-name atlassian -- https://mcp.atlassian.com/v1/mcp

The first run opens a browser for OAuth if no stored credentials exist.

Streamable HTTP frontend¶

mcp-compressor -c medium --transport streamable-http --port 9000 -- python server.py

CLI mode¶

CLI mode writes a shell script that calls the live local session created by mcp-compressor.

mcp-compressor --cli-mode --server-name atlassian -- https://mcp.atlassian.com/v1/mcp

Then run commands through the generated script:

atlassian --help
atlassian get-accessible-atlassian-resources

Use an explicit output directory:

mcp-compressor --cli-mode --server-name alpha --output-dir ./bin -- python server.py

Code Mode¶

Code Mode generates Python or TypeScript functions for backend MCP tools while keeping the local mcp-compressor session alive. By default, generated Code Mode files are written under ./dist in the current working directory.

mcp-compressor --code-mode python --server-name atlassian -- https://mcp.atlassian.com/v1/mcp

mcp-compressor --code-mode typescript --server-name atlassian -- https://mcp.atlassian.com/v1/mcp

See Code Mode and generated clients for examples of generated functions and agent usage.

Just Bash mode¶

mcp-compressor --just-bash-mode --server-name atlassian -- https://mcp.atlassian.com/v1/mcp

Just Bash mode lets language hosts register MCP tools as shell-style commands. See Just Bash.

LLM management¶

mcp-compressor can optionally use a small local LLM for proxy-layer assistance. The llm subcommand manages the required runtime and model assets.

# Check installation status
mcp-compressor llm status

# Download the managed llama-server runtime and default model
mcp-compressor llm pull

# Run a quick inference test
mcp-compressor llm test

# Remove all managed runtime and model assets
mcp-compressor llm remove

All assets are stored locally. Nothing is downloaded unless you explicitly run mcp-compressor llm pull or mcp-compressor llm test. The standard proxy mode (mcp-compressor -c medium -- ...) is unaffected when no LLM assets are installed.

See CLI reference for all options.