MTA Integration Overview

tobira provides plugins for the most popular mail transfer agents. Each plugin queries the tobira API server via HTTP and maps the ML spam score to MTA-native scoring symbols.

Supported MTAs

MTA	Plugin Language	Difficulty	Status
rspamd	Lua	Low	Stable
SpamAssassin	Perl	Medium	Stable
Haraka	Node.js	Low	Stable
Postfix milter	Python	Medium	Stable

How It Works

All MTA plugins follow the same pattern:

1. Intercept — The plugin hooks into the MTA's message processing pipeline
2. Extract — Email body (and optionally headers) are extracted
3. Query — An HTTP POST request is sent to POST /predict on the tobira API server
4. Score — The ML spam probability is mapped to tiered symbols/rules

Score Mapping

All plugins use consistent symbol names and weights:

ML Score	Symbol	Weight	Meaning
>= 0.9	TOBIRA_SPAM_HIGH	+8.0	High confidence spam
>= 0.7	TOBIRA_SPAM_MED	+5.0	Medium confidence spam
>= 0.5	TOBIRA_SPAM_LOW	+2.0	Low confidence spam
< 0.3	TOBIRA_HAM	-3.0	Likely legitimate

Fail-Open Mode

All plugins support fail-open mode: if the tobira API server is unreachable or returns an error, the email is passed through without ML scoring. This ensures mail delivery is never blocked by an ML service outage.

Prerequisites

Before integrating with your MTA, ensure:

1. The tobira API server is running and accessible from the MTA host
2. The /health endpoint returns {"status": "ok"}
3. Network connectivity between MTA and API server (same host or private network recommended)

# Verify API server health
curl http://127.0.0.1:8000/health