MCP Server Integration

Use Papr through MCP so AI clients can store and retrieve memory with consistent tool contracts.

What You Build

A configured MCP server connected to your Papr project
A stable tool surface for memory read and write operations
A repeatable setup for local development and production clients

Prerequisites

Papr API key in PAPR_MEMORY_API_KEY
A client that supports MCP configuration
Access to the Papr MCP package or server repo

Minimal Setup

Install and configure the MCP server.
Add your API key to the MCP runtime environment.
Register the server in your MCP client config.
Restart the client and verify tools are available.

Recommended Tool Contract

Expose a minimal, predictable set of tools:

store_memory
search_memory
compress_session_context
query_graph

Keep parameter names aligned with the API:

external_user_id
memory_policy
enable_agentic_graph

Example Memory Calls

Store memory

{
  "content": "User prefers weekly release summaries by email.",
  "external_user_id": "user_123",
  "memory_policy": {
    "mode": "auto",
    "consent": "implicit",
    "risk": "none"
  }
}

Search memory

{
  "query": "What communication preference does the user have?",
  "external_user_id": "user_123",
  "enable_agentic_graph": true
}

Use response_format=toon when search output is passed directly to an LLM.

Production Settings

Keep tenant scope explicit with organization_id and namespace_id
Set conservative limits (max_memories, max_nodes) for predictable latency
Log tool input and output metadata for traceability
Add retry logic for transient network failures

Validation Checklist

MCP client lists Papr tools after startup
store_memory returns success for a test payload
search_memory returns user-scoped results
Session compression tool returns context_for_llm for long threads

Troubleshooting

If tools fail to load or requests fail at runtime, use Error Playbook and verify MCP runtime environment variables are set correctly.