Other AI Clients
Parametric Memory uses the MCP 2025-03-26 Streamable HTTP transport, which is compatible with any client that supports MCP. Below are setup instructions for the most popular clients.
Connection details
Regardless of client, you need two things from your Dashboard:
| Field | Where to find it |
|---|---|
| MCP URL | https://abc123.parametric-memory.dev/mcp |
| API Key | Dashboard → Settings → API Key (starts with mmk_) |
Cursor
In Cursor, go to Settings → MCP → Add Server and enter:
{
"name": "Parametric Memory",
"type": "streamable-http",
"url": "https://abc123.parametric-memory.dev/mcp",
"headers": {
"Authorization": "Bearer mmk_your_key_here"
}
}Once saved, Cursor's Agent mode will have access to all Parametric Memory tools automatically.
Cline (VS Code extension)
In the Cline extension settings, find MCP Servers and add:
{
"parametric-memory": {
"type": "streamable-http",
"url": "https://abc123.parametric-memory.dev/mcp",
"headers": {
"Authorization": "Bearer mmk_your_key_here"
}
}
}Windsurf
Go to Settings → Cascade → MCP Servers and add a new HTTP server entry with the URL and Bearer token header. Windsurf uses the same Streamable HTTP format as Claude Desktop.
Continue.dev
In your .continue/config.json, add under mcpServers:
{
"mcpServers": [
{
"name": "parametric-memory",
"transport": {
"type": "streamable-http",
"url": "https://abc123.parametric-memory.dev/mcp",
"requestOptions": {
"headers": {
"Authorization": "Bearer mmk_your_key_here"
}
}
}
}
]
}Any custom MCP host
If you are building your own MCP host or agent framework, connect using the standard Streamable HTTP transport:
Transport: Streamable HTTP (MCP 2025-03-26)
URL: https://abc123.parametric-memory.dev/mcp
Auth: Authorization: Bearer mmk_your_key_hereOAuth2 (optional)
Parametric Memory also supports OAuth2 for environments that prefer token exchange over static keys:
POST https://abc123.parametric-memory.dev/oauth/authorize
Body: { "apiKey": "mmk_your_key_here" }
Response: { "access_token": "...", "expires_in": 3600 }Use the returned access_token as the Bearer token in subsequent requests. Tokens expire after 1 hour; re-request as needed.
Available tools
All clients get the same set of tools once connected. See the MCP Tool Reference for the complete list with parameters and examples.
The core tools your AI will use most often are:
memory_session_bootstrap— load relevant context at session startsession_checkpoint— save atoms and relationships during a sessionmemory_search— semantic search across stored atomsmemory_train— reinforce important memory arcs
Testing the connection
Once configured, you can verify the connection by asking your AI:
"Call memory_session_bootstrap with objective: 'test connection' and tell me what it returns."
If the connection is working, you will see a response with the bootstrap payload (likely empty on first use). If you see an error, check the troubleshooting section in the Claude docs — the same steps apply to all clients.