Memory Plane getting started

Connection

MCP endpoint

https://api.mcpkeeper.com/mcp/memory-plane

Generic MCP-over-HTTP config

{
  "mcpServers": {
    "memory-plane": {
      "type": "http",
      "url": "https://api.mcpkeeper.com/mcp/memory-plane"
    }
  }
}

Proof flow

1. 1. Call tools/list to verify the endpoint is reachable and inspect the live tool metadata.
2. 2. Call prepare_memory_context once with a secret context key to get a reusable context ID.
3. 3. Use that context ID on record_memory_event, search_memory, get_context_for_task, and get_memory_timeline.
4. 4. If a tool is paid, an unpaid direct alias call returns 402 Payment Required. Retry with an x402 payment signature and the tool executes.

Build your own agent skill

If you are wiring Memory Plane into your own agent framework, keep the skill small and procedural. The durable pattern is to resume from stored state first, then write a fresh checkpoint before the run stops.

1. 1. Call prepare_memory_context once per stable workflow, tenant, or agent lane, then cache the returned context ID.
2. 2. At the start of a resumed run, call get_context_for_task before taking new actions so the run starts from current memory instead of a transcript recap.
3. 3. Write workflow_state for current truth, checkpoint for compact progress, and handoff before stopping or switching actors.
4. 4. Use stable taskId and workflowId values so resume, supersession, and timeline queries land on the same lane.
5. 5. Use search_memory for ranked recall and get_memory_timeline when chronology matters more than ranking.
6. 6. Do not treat Memory Plane as a generic full-chat log. Store the state, decisions, failures, and next actions that another run will actually need.

Copy-paste client snippets

These are starter instruction snippets you can paste into your client-specific agent rules. Keep them close to the task runner so the agent learns to resume from memory before it writes new state.

Claude

Use MCP Keeper Memory Plane for resumable work.

At the start of a task, if no contextId is known, call prepare_memory_context once and reuse the returned contextId for later runs.
Before doing new work on a resumed task, call get_context_for_task with the stable taskId.
During execution, write record_memory_event entries with entryType=workflow_state for current truth and next action, entryType=checkpoint for compact progress, and entryType=handoff before stopping or transferring work.
Use search_memory for ranked recall and get_memory_timeline when chronological replay matters.
Do not use Memory Plane as a generic transcript dump; store only the state, decisions, failures, and next steps needed by a later run.

Cursor

Memory rule: use MCP Keeper Memory Plane to persist resumable agent state.

1. Bootstrap once with prepare_memory_context if the workflow does not have a contextId yet.
2. Start resumed work with get_context_for_task before taking action.
3. Write workflow_state for current status and next action.
4. Write checkpoint after meaningful progress.
5. Write handoff before ending the run or switching actors.
6. Keep taskId and workflowId stable across retries.
7. Use search_memory for recall and get_memory_timeline for ordered replay.
8. Never store the full chat transcript unless it is directly needed for later execution.

GitHub Copilot

When Memory Plane is available, use it as the continuity layer for multi-step tasks.

If there is no contextId for the workflow, call prepare_memory_context once and reuse the returned contextId.
At the beginning of resumed work, call get_context_for_task with the taskId before issuing new writes.
Use record_memory_event with entryType=workflow_state for current truth, checkpoint for compact progress, episode for notable failures or decisions, and handoff before stopping.
Use search_memory when you need ranked prior context. Use get_memory_timeline when sequence matters.
Prefer concise, high-value state over transcript logging.

Free bootstrap example

curl -sS https://api.mcpkeeper.com/mcp/memory-plane   -H 'content-type: application/json'   -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "prepare_memory_context",
      "arguments": {
        "contextKey": "customer-acme-agent-01-9f3b7d2c",
        "contextType": "agent_workflow",
        "description": "Acme support agent"
      }
    }
  }'

Paid alias route shape

https://api.mcpkeeper.com/mcp/memory-plane/__tool__/record_memory_event