interserve

# interserve

Codex spark classifier and context compression for Claude Code.

## What this does

interserve provides three MCP tools for token-efficient document handling:

**classify_sections**: takes a markdown document and classifies each section into flux-drive review domains (architecture, safety, correctness, etc.) by dispatching a Codex spark. This lets interflux know which review agents to launch without Claude having to read the entire document first.

**extract_sections**: splits a markdown document by `##` headings while properly handling fenced code blocks. Simple structural extraction, no AI involved.

**codex_query**: delegates file reading to Codex to save Claude's context window. When you need information from a large file but don't want to burn context tokens reading it, codex_query reads it in a separate process and returns a summary.

## Installation

First, add the [interagency marketplace](https://github.com/mistakeknot/interagency-marketplace) (one-time setup):

```bash
/plugin marketplace add mistakeknot/interagency-marketplace
```

Then install the plugin:

```bash
/plugin install interserve
```

Requires Clavain's `dispatch.sh` for Codex spark dispatch (set `INTERSERVE_DISPATCH_PATH`).

## Architecture

```
cmd/interserve-mcp/    Go MCP server (mark3labs/mcp-go)
bin/launch-mcp.sh      Server launcher
```

Go binary with stdio MCP transport. Starts on-demand, no systemd unit.