Claude Code Memory vs ProductLobster
Anthropic now ships project memory natively in Claude Code: CLAUDE.md captures human-authored project rules; MEMORY.md captures Claude-authored build commands, debugging lessons, and workflow habits. Both load hierarchically from the file system. Zero third-party API calls.
Auto Memory solves the engineer's context problem. ProductLobster solves the product organization's problem. They compose architecturally — but they answer different questions for different people.
How they differ.
| Attribute | Claude Code Memory | ProductLobster |
|---|---|---|
| What it remembers | Build commands, debugging lessons, workflow habits | Customer signals, competitive positioning, strategic decisions, hypotheses, experiments, roadmap state, outcomes |
| Ontology | Markdown files, free-form text | Typed: problem, customer, competitor, opportunity, decision, hypothesis, experiment, roadmap, outcome — each with a citation to the source and a timestamp showing how recent it is |
| Who it's for | Engineers using Claude Code | Founders and operators running product work without a PM team |
| Workflow | None — memory is read into the engineer's context window | Opinionated PM workflow: decision checkpoints, refinement chats, structured analytical pipelines, prototype iteration |
| Methodology | Whatever the user writes into their notes | The work a senior PM team would do, applied invisibly: customer research, competitive scan, strategic framing, prototype iteration |
| Composability | File system, hierarchical CLAUDE.md + MEMORY.md | Native MCP server; Claude Code can read ProductLobster's Product Brain via MCP, ProductLobster can read Auto Memory via the file system |
| Price | Free with Claude Pro ($20/mo) | Free during beta. $29/mo Plus when billing opens. |
When each is the right tool.
Claude Code Memory is right when…
- You're an engineer who wants Claude Code to remember your project conventions, build commands, and debugging notes across sessions
- You don't need a structured product ontology — text notes in Markdown files are enough
- You want zero third-party services in the loop and zero ongoing subscription cost beyond Claude Pro
ProductLobster is right when…
- You need to run product work — customer research, competitive analysis, strategic framing, prototype iteration — and have it compound across analyses
- You want the discipline of senior PM workflow applied invisibly: decision checkpoints, evidence confidence tagging, structured option capture
- You need typed product memory (problem, customer, competitor, opportunity, decision, hypothesis, experiment, roadmap, outcome), each fact citing its source and showing how recent it is — not text blobs
- You want to query your Product Brain from Claude Code itself via MCP (coming soon) — keeping Auto Memory for engineering work and ProductLobster for product work, both surfaces composing
How they compose architecturally.
Auto Memory and ProductLobster aren't opposed. They ride the same architectural pattern — file-system-as-state, composable via MCP — and they answer different layers of the same operator's day.
ProductLobster can read your CLAUDE.md and MEMORY.md via the file system. Claude Code will be able to read your Product Brain via MCP once that surface ships (coming soon). Engineers keep Auto Memory for the build context. Product people use ProductLobster for the product context. Both surfaces compose; neither replaces the other.
Memory is the substrate, shipped everywhere. PM-workflow ontology on top of it is the irreducible difference.