Data Storage

Local Storage

By default, summaries are stored in ContextPool/ inside your project directory:

your-project/
└── ContextPool/
    └── fetched/
        └── 2026-03-27T08-00-00Z/
            ├── abc123.summary.md    ← full insights for one session
            └── index.json           ← metadata index for fast lookup

Each .summary.md contains structured insights in this format:

# Session abc123

- **bug** ESM import fails silently in test runner — add "type": "module" to package.json
- **gotcha** reqwest needs rustls-tls feature on Alpine/ARM — default OpenSSL fails to link
- **decision** chose SQLite over Postgres — latency requirements don't justify the ops overhead

index.json stores metadata (ids, titles, types, tags, timestamps) for fast retrieval without reading every summary file.

---

Storage Locations

Flag / Command	Where summaries are written
cxp init --local	./ContextPool/ (next to your code)
cxp init --out <dir>	<dir>/
cxp init (no flag)	OS app data directory
cxp mcp --data-dir <dir>	<dir>/ (MCP server reads from here)
cxp export --out <dir>	<dir>/

The --local flag is recommended for most projects — it keeps memory next to the code and makes it easy to commit to git.

---

Team Cache

After cxp pull, team insights are cached at:

~/.cache/contextpool/team-cache/<project-name>/team-insights.md

The MCP server's search_context and get_project_context tools automatically include these alongside local summaries.

---

Committing to Git

Committing your ContextPool/ directory is a great way to share memory with teammates without setting up Team Sync:

# Add to git
git add ContextPool/
git commit -m "chore: add contextpool memory"

# Or add to .gitignore if you prefer local-only
echo "ContextPool/" >> .gitignore

When a teammate clones the repo, they immediately have your insights available in their MCP sessions.

---

Disk Usage

Each .summary.md file is typically 500–2,000 bytes. A project with 100 sessions produces roughly 50–200 KB of summary files. The overhead is negligible.