Measure what context compression actually costs an LLM agent: where a session's tokens go, and whether a rewrite pays for the prompt cache it invalidates. Rust crate, MCP server and benchmark harness.
Setup
Choose your client, copy the config, then fill in any required environment variables.
claude mcp add distil-munhq -- npx -y @munhq/distil~/.codex/config.toml
[mcp_servers.distil-munhq]
command = "npx"
args = ["-y", "@munhq/distil"]
# codex mcp add distil-munhq -- npx -y @munhq/distil~/.cursor/mcp.json
{
"mcpServers": {
"distil-munhq": {
"command": "npx",
"args": [
"-y",
"@munhq/distil"
]
}
}
}.vscode/mcp.json
{
"servers": {
"distil-munhq": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@munhq/distil"
]
}
}
}~/.gemini/settings.json
{
"mcpServers": {
"distil-munhq": {
"command": "npx",
"args": [
"-y",
"@munhq/distil"
]
}
}
}claude_desktop_config.json
{
"mcpServers": {
"distil-munhq": {
"command": "npx",
"args": [
"-y",
"@munhq/distil"
]
}
}
}.trae/mcp.json
{
"mcpServers": {
"distil-munhq": {
"command": "npx",
"args": [
"-y",
"@munhq/distil"
]
}
}
}