Afterpaths

What is Afterpaths?

Afterpaths captures sessions across AI coding tools (Claude Code, Cursor, Codex CLI), summarizes them, and automatically extracts rules—so your agents learn from every session and you retain rich context. It includes a Model Context Protocol (MCP) server that exposes session history, summaries, and rules as native tools for AI assistants.

How to use Afterpaths?

Install with pip install afterpaths (or pip install afterpaths[mcp] for MCP support). Navigate to your project and run ap audit to see an overview. Recommended flow: ap log to browse sessions, ap summarize <ref> (requires ANTHROPIC_API_KEY), ap rules to extract rules into .claude/rules/, ap search <query> to find past discussions, and ap stats to track performance. For the MCP server, configure it with claude mcp add afterpaths -- afterpaths-mcp or run python -m afterpaths.mcp_server. Both afterpaths and ap work identically.

Key features of Afterpaths

• Unified session view across all AI coding tools
• LLM-generated session summaries (discoveries, dead ends, decisions)
• Automatic rule extraction into .claude/rules/ and .cursor/rules/
• Cross-session search (ap search) with optional deep search
• Session analytics: tokens, errors, daily trends
• MCP server with 6 native tools (list, show, search, summarize, artifacts, rules)

Use cases of Afterpaths

• Compare model performance across tools to pick the best model for your stack
• Capture and reuse discoveries as rules to prevent repeated mistakes
• Search past sessions for context when resuming work after days or weeks
• Track ongoing session metrics to identify expensive or ineffective approaches
• Retain session history that would otherwise be auto-deleted after 30 days

FAQ from Afterpaths

What does Afterpaths do that my IDE doesn't?

Afterpaths gives a single view across Claude Code, Cursor, and Codex CLI—something no single IDE provides. It automatically extracts rules from sessions and surfaces comparative model performance.

What dependencies or API keys are required?

Audit, log, and search commands work without any API key. Summarization requires an ANTHROPIC_API_KEY. The MCP server uses local files only.

Where do summaries and rules live?

Summaries are stored in .afterpaths/ (gitignored). Rules are written to .claude/rules/ and .cursor/rules/ (these should be committed). Everything stays in your project; no data leaves your machine except when sending session text to the LLM for summarization.

Does Afterpaths support all Cursor sessions?

Cursor support is partial—the adapter reads several known formats from state.vscdb, but Cursor changes its storage schema often. Some sessions may show Entries: 0. In that case, ap show <ref> --raw emits a warning; you can still access the raw file directly with sqlite3. Tool provenance features only cover sessions whose entries the adapter can read.

How are manually written rules preserved?

Each rule file has <!-- afterpaths:manual:start --> and <!-- afterpaths:manual:end --> markers. Content between those markers survives every ap rules run. Auto-generated content is regenerated. Content outside markers (other than boilerplate) blocks the run unless --force is used.