MCP Server

Remind can run as an MCP (Model Context Protocol) server, allowing AI agents in IDEs like Cursor to use it as their memory system over HTTP.

When to use MCP

MCP is a good fit when:

• Your IDE/agent supports MCP natively (Cursor, Claude Desktop, etc.)
• You want centralized memory across projects (named stores at ~/.remind/<name>.memory)
• You want the agent to use Remind tools directly without shell access

For project-scoped memory with composable workflows, see Skills + CLI instead.

Starting the server

# After pip install
remind-mcp --port 8765

# Or with uv (no install needed)
uvx remind-mcp --port 8765

# Or with Docker
docker compose up -d

Connecting from Cursor

Add to your project's .cursor/mcp.json:

{
  "mcpServers": {
    "remind": {
      "url": "http://127.0.0.1:8765/sse?db=my-project"
    }
  }
}

The db parameter is a simple name that resolves to the Git-friendly file store ~/.remind/{name}.memory. Each project can have its own store. Explicit SQL URLs, including sqlite:////absolute/path/to/remind.db, continue to select SQL storage.

Available tools

Full parameter lists and examples: MCP tools reference.

Summary:

Tool	Purpose
remember	Store an episode (returns collision info for facts)
remember_batch	Atomically capture multiple separate assertions
recall	Spreading activation (+ optional entity, labels, as_of)
snapshot	Batch read (pending episodes, conflicts, entities, labels, concepts)
apply	Batch write (create concepts, supersede facts, resolve conflicts, label/unlabel)
poll_events	Poll durable changes with a generation-qualified cursor
events	Legacy numeric-sequence change feed for one projection
doctor	Diagnose integrity, embedding compatibility, and source freshness
stats	Memory statistics
list_conflicts / resolve_conflict / dismiss_conflict	Manage detected conflicts
inspect	Concept detail with facts and supersession history
entities / inspect_entity	List and inspect entities
update_episode / delete_episode / restore_episode	Manage episodes
update_concept / delete_concept / restore_concept	Manage concepts
list_deleted	Review soft-deleted episodes/concepts

See MCP Tools reference for the full parameter list of every tool.

Agent instructions

Copy docs/AGENTS.md into your project to instruct AI agents how to use Remind's MCP tools. This covers:

• Capture — When and how to use remember (including fact collisions)
• Recall — Semantic, entity, and label-scoped queries with time-travel
• Curation — Using snapshot and apply for batch memory management

Docker

Run Remind as a persistent background service:

docker compose up -d

The container:

• Mounts ~/.remind from your host for database persistence
• Exposes port 8765 for MCP SSE, Web UI, and REST API
• Uses local embeddings by default (no API keys needed)
• Restarts automatically on crash/reboot

Access endpoints:

• MCP SSE: http://localhost:8765/sse?db=my-project
• Web UI: http://localhost:8765/ui/?db=my-project
• REST API: http://localhost:8765/api/v1/...

Optional: Remote embeddings

To use OpenAI embeddings instead of local:

cp .env.example .env   # Edit with OPENAI_API_KEY
docker compose up -d

Or set environment variables when running the server directly.