MongoDB

When to Use

• Querying conversations, messages, or agents in LibreChat’s MongoDB
• Any script that needs pymongo in the Railway environment

Connection

from pymongo import MongoClient

def get_env(key):
    with open("/proc/1/environ", "rb") as f:
        for entry in f.read().decode("utf-8", errors="replace").split("\0"):
            if entry.startswith(f"{key}="):
                return entry.split("=", 1)[1]
    raise RuntimeError(f"{key} not found in /proc/1/environ")

MONGO_URI = get_env("MONGO_URI")
db = MongoClient(MONGO_URI)["test"]

Python Packages in Railway

Packages are baked into the Docker image via Forge/Configs/LibreChat_Service/Dockerfile.librechat. To add a new package, add it to the pip install line in the Dockerfile and redeploy.

Installed packages:

• pymongo — MongoDB driver
• dnspython — pymongo dependency

Collections & Schemas

conversations

Field	Type	Description
conversationId	string (UUID)	Primary identifier
title	string	Conversation title
messages	list[ObjectId]	References to messages collection
updatedAt	datetime	Last update timestamp
user	ObjectId	User who owns the conversation

messages

Field	Type	Description
messageId	string (UUID)	Primary identifier
conversationId	string (UUID)	Parent conversation
sender	string	"User" or agent display name
isCreatedByUser	bool	True for user messages
text	string	Plain text of the message (may be empty for tool-heavy turns)
content	list[dict]	Structured content items (see below)
tokenCount	int	Token count for the message
createdAt	datetime	Creation timestamp

content item types

text: {"type": "text", "text": "..."}

tool_call:

{
  "type": "tool_call",
  "tool_call": {
    "id": "call_...",
    "name": "tool_name",
    "args": "{\"param\": \"value\"}",
    "output": "result text",
    "progress": 1
  }
}

agents

Field	Type	Description
id	string	Agent ID (agent_...)
name	string	Display name
instructions	string	System prompt / instructions
model_parameters	dict	Model-specific settings
tools	list	Enabled tools

Rules

• Never expose MONGO_URI in conversation output — redact in transcripts
• MONGO_URI is read from /proc/1/environ — never hardcode connection strings
• Always close connections or use context managers for scripts that run standalone