Other MCP clients

Connect any MCP client that supports Streamable HTTP to Lore's memory tools. The client gets the memory tools, but Lore doesn't load startup memory or recall for it, so the agent needs instructions to call lore_boot and lore_search itself.

EndpointCopy Anchor Link

Point the client at <server>/api/mcp?client_type=mcp. Lore serves MCP over Streamable HTTP. The client sends JSON-RPC requests with POST, and Lore answers with JSON. Lore doesn't issue MCP session IDs.

Keep client_type=mcp in the URL. Lore uses it to show activity from generic MCP clients separately in Recall analytics.

Example configCopy Anchor Link

Each client has its own configuration format, so treat this JSON as an example only. Use the keys your client expects.

{
  "mcpServers": {
    "lore": {
      "type": "http",
      "url": "https://api.loremem.com/api/mcp?client_type=mcp",
      "headers": {
        "Authorization": "Bearer lm_your_token"
      }
    }
  }
}

What you getCopy Anchor Link

• The memory tools: lore_guidance, lore_status, lore_boot, lore_get_node, lore_search, lore_list_domains, lore_create_node, lore_update_node, lore_delete_node, and lore_move_node. The tools reference lists the inputs of each tool.
• Usage guidance. When the server has guidance text, Lore sends it as MCP server instructions in the initialize response. Clients that pass server instructions to the model give the agent that guidance.

What you don't getCopy Anchor Link

• Startup memory. The client doesn't call lore_boot by itself. lore_boot returns the global boot nodes, core://agent, core://soul, and preferences://user. A generic MCP client has no runtime boot node of its own.
• Recall. No <recall> block is added to prompts.
• Skills. The endpoint doesn't serve lore_skill_* tools. Skill tools come only from the runtime integrations.

To close the gap, add these instructions to your client's system prompt:

At the start of each session, call lore_boot and read the result before you answer.
Before you rely on a remembered fact, call lore_search, then open the best match with lore_get_node.

Without these instructions, the agent reads Lore only when you ask it to.

VerifyCopy Anchor Link

1. Call lore_status from the client. It returns the server's health.
2. Call lore_boot. The result includes core://agent, core://soul, and preferences://user.
3. Call lore_search with a term from a memory you have. The result lists the matching node.

If the client can't connect, or it gets 401 Unauthorized, see Agent connection issues.