Getting Started

Build with Coding Agents

Let Claude Code, Cursor, Codex, or any AgentSkills-compatible agent build your MemorySync integration — correctly, first try. The building-with-memorysync skill supplies the decision rules, the zero-auth documentation MCP server supplies always-current API facts, and self-serve evaluation keys let the agent prove the integration works before anyone signs up.

The three pieces

Piece	What it does
The `building-with-memorysync` skill	A decision layer, not an API reference: how to scope projects and end users, choose ingest and retrieval paths, evaluate honestly, and which adapter to delegate to. Volatile facts deliberately live elsewhere — if the skill and the live docs disagree, the live docs win.
The docs MCP server	`https://docs.memorysync.io/mcp`, no auth: `search_docs`, `read_doc` (whole pages — preferred), `list_doc_sections`, the `implement_with_memorysync` prompt, and the `docs-index` resource (llms.txt). The agent reads real pages instead of inventing endpoints.
Self-serve evaluation keys	`POST /evaluation/keys` mints a working, metered, expiring key with no human signup — the agent can run an add → query → assert round-trip and report proof before you create an account. No other memory platform offers this.

	Mem0	Supermemory	Zep	MemorySync
Builder skill ships	✓ 6-skill graph	✓ + 78KB references	✓ decision layer	✓ decision layer
Skill can go stale	△ llms.txt fetch pattern	❌ inline endpoints in the skill body	✓ docs-MCP authority	✓ docs-MCP authority + CI pins every named tool/endpoint/guide to the codebase
Agent can self-provision a key	✗ human signs up	✗ OAuth human flow	✗ enterprise IdP	✓ `POST /evaluation/keys`, autonomous
Quota behaviour taught	✗	✗	✗	✓ strict-429 eval vs silent production, exact payloads
Docs MCP surface	hosted, needs API key	docs MCP (separate page)	1 search tool + resources	✓ 3 tools + prompt + resources, zero-auth

Install

# Marketplace plugin (skill + docs MCP together):
/plugin marketplace add Rafay121/memorysync-plugins
/plugin install building-with-memorysync@memorysync

The docs MCP server needs no authentication — it serves documentation, not memories. Pair it with the skill for best results: the skill decides, the docs server informs.

The starter prompt

You are integrating MemorySync into my app.
- Use the memorysync-docs MCP server (or https://docs.memorysync.io/llms.txt)
  before inventing any endpoint or parameter.
- Follow the building-with-memorysync skill for scoping, ingest, retrieval,
  and the evaluation loop.
- Prove the integration with a minted evaluation key before asking me for one.

With the skill installed, the agent already knows the workflow — the prompt just sets expectations. Without the skill, the prompt alone still steers the agent to the docs MCP and llms.txt.

The autonomous evaluation loop

Step	What the agent does	Expected
1. Mint	`POST /evaluation/keys` (body optional: `{"agent_caller": "claude-code"}`)	`201` with `api_key`, `default_user_id`, `mcp_url`, `expires_at`, `limits`
2. Round-trip	Add one fact, query it back with the same `X-End-User-ID`	The stored text returns
3. Idempotency proof	Send the identical `add_turn` twice	The second response reports the row already exists — no duplicate
4. Meter check	`GET /evaluation/usage`	The adds/retrievals the loop consumed
5. Graduate	The human claims the account or creates a dashboard key	Same API, production silent-mode quota, higher limits

Supported versions

Surface	Requires	Verified on
`building-with-memorysync` skill 1.0.0	Any AgentSkills 1.0 client (Claude Code, Cursor, Copilot, OpenCode, …)	9 CI contract checks: spec compliance, and every docs-MCP tool, REST endpoint, and guide slug the skill names verified against the actual codebase — the skill cannot silently go stale
Builder plugin (skill + docs MCP)	Claude Code 2.x	`claude plugin validate --strict` on the plugin and the marketplace root

Where to go next

MemorySync MCP

The memory server your built integration will use.

Quickstart

The human-paced version of what the agent automates.

Agent Memory

What an agent should and should not store.

Was this page helpful?

← Previous

Agent Memory

RAG & Knowledge Assistants