Claude Code Memory Demo: Record, Recall, Show Lineage

The shortest believable Memnode demo

If Memnode is going to earn attention, the first public artifact cannot be abstract product copy.

It has to show one concrete loop that a developer can run and verify:

1. install Memnode into Claude Code
2. teach the agent one repo convention
3. ask again in a later session
4. inspect the lineage so the memory is not a black box

That loop is the wedge.

Install it into Claude Code

From the repo root:

cargo build --release --manifest-path data-plane/Cargo.toml

./data-plane/target/release/memnode install-mcp   --target claude-code   --verify

--verify matters. It drives the installed entry the way the MCP client would, so the setup fails early if the command path is wrong.

Teach one repo convention

Use a convention that actually matters in later sessions.

For example:

remember that this repo uses Zod for request validation in the route layer

The point is not to save a random sentence. The point is to store something the agent will probably need again.

Recall it later

Now ask the question the way it would naturally appear during real work:

what do we use to validate request payloads?

The useful answer is not just the fact. The useful answer is the fact plus enough context to trust it.

Expected shape:

This repo uses Zod in the route layer for request validation.

Show lineage

This is the part most memory products skip.

Ask:

show lineage for that memory

Expected shape:

observed from repo_convention
anchored to app/api/users/route.ts
available across later sessions

That is the public proof artifact. The agent did not just retrieve something fuzzy. It can show what it saw and why it recalled it.

Why this demo works

• The pain is immediate.
• The setup is short.
• The proof is visible.
• The workflow already exists inside Claude Code.

This is why coding assistants are the first acquisition funnel even though Memnode is not limited to coding use cases.

What this means for support or internal agents

The same loop applies outside coding:

• record a support policy
• recall it in a later customer conversation
• inspect the lineage
• correct it when the policy changes

That is why support agents stay in product scope even if the first public demo is coding-first.

Why this is better than black-box memory

The real problem with agent memory is not storing more text.

The real problem is operating it:

• what did the agent remember?
• why did it remember it?
• is it stale?
• what replaced it?

Memnode is trying to make that memory inspectable instead of magical.

Next step

If this demo makes sense, the next page is the install surface itself: Open the MCP install page.