Skip to main content
Glama

Memory Interchange Format (MIF)

PyPI npm License Tests Docs

Your AI agent has 6 months of memories in System A. You want to try System B. Without MIF, you lose everything. With MIF:

pip install mif-tools
mif convert mem0_export.json --to shodh -o memories.mif.json

Done. Your memories are portable.

What is MIF?

A vendor-neutral JSON envelope for AI agent memories. Like vCard for contacts or iCalendar for events — a minimal schema so memories move between providers without data loss.

3 required fields. That's it.

{
  "mif_version": "2.0",
  "memories": [
    {
      "id": "a1b2c3d4-e5f6-7890-abcd-ef1234567890",
      "content": "User prefers dark mode across all applications",
      "created_at": "2026-01-15T10:30:00Z"
    }
  ]
}

Everything else — memory types, tags, entities, embeddings, knowledge graph, vendor extensions — is optional. Add what you have, ignore what you don't.

Related MCP server: CarpeOS MCP Server

Install

# Python
pip install mif-tools              # core (zero dependencies)
pip install mif-tools[validate]    # with JSON Schema validation
pip install mif-tools[mcp]         # with MCP server

# Node.js / TypeScript
npm install @varunshodh/mif-tools

Convert Between Formats

# mem0 → MIF
mif convert mem0_export.json --from mem0 -o memories.mif.json

# MIF → Markdown (Obsidian/Letta style)
mif convert memories.mif.json --to markdown -o memories.md

# Auto-detect source format
mif convert any_memory_file.json -o output.mif.json

# Inspect any memory file
mif inspect memories.json

# Validate MIF document
mif validate memories.mif.json

Python API

from mif import load, dump, convert, MifDocument, Memory

# Load from any format (auto-detects mem0, markdown, generic JSON, MIF)
doc = load(open("mem0_export.json").read())
print(f"{len(doc.memories)} memories loaded")

# Convert between formats in one line
markdown = convert(data, from_format="mem0", to_format="markdown")

# Create memories from scratch
doc = MifDocument(memories=[
    Memory(
        id="123e4567-e89b-12d3-a456-426614174000",
        content="User prefers dark mode",
        created_at="2026-01-15T10:30:00Z",
        memory_type="observation",
        tags=["preferences", "ui"],
    )
])
print(dump(doc))  # MIF v2 JSON

# Deep validation (UUIDs, references, timestamps, embedding dimensions)
from mif import validate_deep
ok, warnings = validate_deep(open("export.mif.json").read())

Add MIF to Your MCP Server (10 lines)

from mif import load, dump

# Export handler
def export_memories(user_id: str) -> str:
    memories = my_storage.get_all(user_id)
    return dump(memories)

# Import handler — auto-detects mem0, markdown, generic JSON, MIF
def import_memories(data: str) -> dict:
    doc = load(data)
    for mem in doc.memories:
        my_storage.save(mem.id, mem.content, mem.created_at)
    return {"memories_imported": len(doc.memories)}

Supported Formats

Format

ID

Auto-detect

Description

MIF v2

shodh

"mif_version" in JSON

Native format, lossless round-trip

mem0

mem0

JSON array with "memory" field

mem0 memory exports

CrewAI

crewai

JSON array with "task_description"

CrewAI LTMSQLiteStorage exports

LangChain

langchain

JSON array with "namespace" + "value"

LangChain/LangMem Item format

Generic JSON

generic

JSON array with "content" field

Any JSON memory array

Markdown

markdown

Starts with ---

YAML frontmatter (Letta/Obsidian style)

Full Spec

MIF supports optional fields for rich memory data:

  • Memory typesobservation, decision, learning, error, context, conversation, and custom types

  • Entity references — named entities with type and confidence

  • Embeddings — model name, dimensions, vector (reuse or regenerate)

  • Knowledge graph — entities and relationships with confidence scores

  • Vendor extensions — system-specific metadata preserved on round-trip

  • Privacy — PII detection and redaction markers

Full specification: spec/mif-v2.md | JSON Schema: schema/mif-v2.schema.json

MCP Server

Expose MIF tools to any MCP-compatible AI client:

pip install mif-tools[mcp]
mif mcp

Tools: export_memories, import_memories, validate_memories, inspect_memories, list_formats

Adapters & Implementations

System

Status

Type

shodh-memory

Production

Built-in HTTP API (/api/export/mif, /api/import/mif)

mif-tools (PyPI)

Production

Python package with CLI + MCP server

@varunshodh/mif-tools (npm)

Production

TypeScript/Node.js package with CLI

mem0

Adapter ready

Python + npm

CrewAI

Adapter ready

Python + npm

LangChain

Adapter ready

Python + npm

Generic JSON

Adapter ready

Python + npm

Markdown (YAML frontmatter)

Adapter ready

Python + npm

Design Principles

  1. Minimal — 3 required fields. Everything else is optional.

  2. Extensible — Unknown fields and vendor extensions MUST be preserved on round-trip.

  3. Vendor-neutral — The schema doesn't favor any implementation.

  4. Forward-compatible — Importers MUST ignore unknown fields.

Contributing

We welcome adapter implementations for any memory system. See CONTRIBUTING.md.

License

Apache 2.0

A
license - permissive license
Not graded
quality - not tested
C
maintenance

Maintenance

0dRelease cycle
2Releases (12mo)

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Connectors

  • XMemoOAuth

    Shared, governed long-term memory for AI agents across tools and sessions via MCP and REST.

  • Manage portable AI agent playbooks, Agent Skills, MCP configurations, personas, and memory.

  • Portable memory for AI agents: capture once, recall across Claude, Cursor, and any MCP client.

  • Persistent, inspectable memory for AI agents with lineage, correction, and a hosted MCP endpoint.

View all MCP Connectors

Related MCP Servers

  • F
    license
    Not graded
    quality
    C
    maintenance
    Provides persistent memory tools (recall, remember, checkpoint) for AI agents, enabling them to save and restore state across sessions.
    2
  • A
    license
    Not graded
    quality
    A
    maintenance
    Enables AI agents to capture, search, and manage structured memory from agent sessions with append-only events and provenance tracking, providing eight local MCP stdio tools.
    Apache 2.0
  • A
    license
    Not graded
    quality
    B
    maintenance
    Provides portable memory for AI agents using plain Markdown files. Enables storing, recalling, and managing memories via MCP tools like recall, remember, forget, list, and get.
    18
    1
    MIT
  • A
    license
    Not graded
    quality
    B
    maintenance
    Enables AI agents to store, search, assemble, and manage local-first memories through seven MCP tools, including conversation turns, feedback, status, and dashboard access without cloud dependencies.
    AGPL 3.0

View all related MCP servers

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/varun29ankuS/mif-spec'

If you have feedback or need assistance with the MCP directory API, please join our Discord server