MCP Servers
MemorySync runs two hosted Model Context Protocol servers. One gives an assistant your memory, the other gives a coding agent the current documentation. Both are remote, so there is nothing to run locally.
Two servers, two jobs
The Model Context Protocol is an open standard that lets an AI client call external tools. MemorySync publishes two MCP servers. They solve different problems and you can connect to either or both.
| Server | URL | Auth | Exposes |
|---|---|---|---|
| Memory | https://mcp.memorysync.io/mcp | Browser sign-in, or an API key | 15 tools, 4 resources, 2 prompts |
| Documentation | https://docs.memorysync.io/mcp | None | 3 tools, 2 resources, 1 prompt |
What an assistant gains
Without memory, every conversation starts from nothing: you re-explain your stack, your preferences and your decisions to each new session, in each separate tool. Connecting the memory server means an assistant can look up what you already told it, and record what is worth keeping, on its own initiative.
- Context carries over
- A preference recorded in Claude is available in Cursor, because both read the same memory rather than their own local history.
- The assistant decides when
- You do not call an API. The model calls search_memory before answering and add_memory when you state something lasting.
- Conflicts are visible
- When two memories disagree, get_related_memories follows the recorded contradiction edge so the assistant can ask which one still holds instead of averaging them.
- Deletes are deliberate
- The destructive tools preview by default and act only on an explicit confirmation, so a vague "clear that" cannot erase a scope.
- Your sources are visible
- Content synced from GitHub, Slack, Notion, Google Drive, OneDrive, Granola, Amazon S3 or a crawled site is memory the assistant can cite and explain.
The documentation server
A coding agent writing MemorySync code from training data invents method names, guesses at parameters and misses limits. That is not something it can reason its way out of — the correct answer is in the documentation and nowhere else. The second server puts the documentation inside the agent's reach.
https://docs.memorysync.io/mcp
| Tool | What it does |
|---|---|
search_docs | Search every documentation page and get back matching pages with their headings and a link to the plain Markdown text. |
read_doc | Read one page in full. Only pages in the generated index can be read, so it cannot fetch arbitrary URLs through the server. |
list_doc_sections | List the top-level sections and how many pages each contains, to orient before searching. |
No account, no key, no consent screen. It also publishes an implement_with_memorysync prompt that produces the search-then-read sequence for a given task, so the agent reads before it writes.
Where to go next
If you just want it working, run the installer and restart your client.
npx -y memorysync-mcp-install
- Connect a client
- Per-client configuration, how sign-in works, API keys for headless use, and what to check when a client will not connect.
- Memory tools
- Every tool, what it takes and returns, the resources and prompts, and the two scopes that decide what a connection may do.
- SDKs
- Building the integration yourself instead of delegating it to an assistant.