@ghostpaw/email

A standalone email engine for Node.js, built on SQLite.
Email treats IMAP sync, SMTP send, local caching, threading, search, and
transport errors as one coherent model instead of a grab bag of protocol
adapters. It ships as a single prebundled blob with zero runtime dependencies,
designed for three audiences at once: human developers working directly in
code, LLM agents operating through a structured soul / tools / skills
runtime, and agent operators who want a ready-to-use email CLI binary
without writing any wiring code.
MCP server
Every client below starts the same stdio server: npx -y @ghostpaw/email mcp.
Cached messages live in ~/.ghostpaw/email.db unless the GHOSTPAW_EMAIL_DB
environment variable (absolute path) or --db <path> says otherwise.
Claude Code
claude mcp add ghostpaw-email -- npx -y @ghostpaw/email mcp
or as a plugin, which also installs the skill:
/plugin marketplace add GhostPawJS/email
/plugin install ghostpaw-email@ghostpaw-email
Email CLI (~/.email/config.toml)
[mcp_servers.ghostpaw-email]
command = "npx"
args = ["-y", "@ghostpaw/email", "mcp"]
Cursor (.cursor/mcp.json), Claude Desktop (claude_desktop_config.json)
or a project .mcp.json:
{
"mcpServers": {
"ghostpaw-email": {
"command": "npx",
"args": ["-y", "@ghostpaw/email", "mcp"]
}
}
}
Claude Desktop users can also install server.mcpb from the
latest GitHub release.
A Dockerfile is included for container hosts:
docker build -t ghostpaw-email . && docker run -i --rm -v email:/data ghostpaw-email.
The server speaks MCP 2026-07-28 (stateless, server/discover) and the
initialize-based revisions 2025-11-25, 2025-06-18, 2025-03-26 and 2024-11-05.
Network operations use the existing CLI account configuration (email account add) or
EMAIL_IMAP_HOST, EMAIL_SMTP_HOST, EMAIL_USER and EMAIL_PASS (or
EMAIL_ACCESS_TOKEN). Discovery and cached reads need no account or credentials.
The surface database defaults to ~/.ghostpaw/email.db; point --db at an
existing account cache to read it. Each network call opens and closes its IMAP
session. Connection commands do not retain a session between calls.
Tool CLI
npx -y @ghostpaw/email tools
npx -y @ghostpaw/email call mail_read '{"view":"folders"}' --db :memory:
npx -y @ghostpaw/email call mail_search '{"query":"invoice"}'
Both email and ghostpaw-email run the same CLI. Existing account, read,
search, compose, organize and sync commands remain available. The full guide
is preserved in skills/email/references/cli.md.
Agent Skill: SKILL.md and skills/email/ ship in the package; print the skill
with npx -y @ghostpaw/email skill.
Install
npm install @ghostpaw/email
npm install -g @ghostpaw/email
npx @ghostpaw/email --help
Requires Node.js 24+ (uses the built-in node:sqlite module).
Quick Start
import { Mailbox } from '@ghostpaw/email';
const mailbox = new Mailbox({
imap: { host: 'imap.example.com', port: 993, tls: true },
smtp: { host: 'smtp.example.com', port: 465, tls: true },
auth: { user: 'user@example.com', pass: 'secret' },
storage: 'mail.db',
});
await mailbox.connect();
const folders = mailbox.read.folders();
await mailbox.network.sync();
const messages = mailbox.read.messages('INBOX', { limit: 20 });
await mailbox.write.send({
to: [{ address: 'bob@example.com' }],
subject: 'Hello',
text: 'Sent from @ghostpaw/email.',
});
await mailbox.disconnect();
connect() opens the SQLite database, initialises the schema, upserts the
account, and authenticates the IMAP session — one call does everything.
The Model
Seven stored concepts and three derived ones:
| Account | One email identity with IMAP/SMTP connection config |
| Folder | One IMAP mailbox path with sync metadata and role detection |
| Message | A locally cached email header record with flags, labels, and envelope |
| Body | The decoded text/html content and raw RFC 2822 bytes of one message |
| Attachment | A decoded MIME part with on-demand binary fetch |
| SyncLog | A record of one completed sync pass per folder |
| Thread | A JWZ conversation tree derived from References/In-Reply-To headers |
The model means each kind of mail state has its own home:
| An email inbox | A Folder with role: 'inbox' and synced messages |
| A conversation | A Thread computed from References and In-Reply-To chains |
| An unread email | A Message without \Seen in its flags array |
| A starred email | A Message with \Flagged in its flags |
| A PDF attached to a message | An Attachment with metadata from BODYSTRUCTURE and data fetched on demand |
| "Sync found 12 new messages" | A SyncLog row recording strategy, counts, and duration |
State is derived, not toggled. Thread IDs are computed from header chains at
sync time, folder roles are detected from IMAP special-use attributes, and
attachment metadata is extracted from BODYSTRUCTURE — all materialised during
sync, never stored as manual status fields.
Two Audiences
Human developers
Use the Mailbox class with typed read, write, and network surfaces for
direct-code access to every mail operation:
import { Mailbox } from '@ghostpaw/email';
const mailbox = new Mailbox(config);
await mailbox.connect();
const threads = mailbox.read.threads('INBOX', { limit: 20 });
const detail = await mailbox.read.getMessage('INBOX', uid);
await mailbox.write.reply('INBOX', uid, { text: 'Acknowledged.' });
await mailbox.write.archive('INBOX', [uid]);
See HUMAN.md for the full human-facing guide with worked
examples.
LLM agents
Use the tools, skills, and soul namespaces for a structured runtime
surface designed to minimise LLM cognitive load:
import { skills, soul, tools } from '@ghostpaw/email';
const allTools = tools.aligned.emailTools;
const readTool = tools.aligned.getEmailToolByName('mail_read');
const triageSkill = skills.getEmailSkillByName('triage-inbox');
const systemPrompt = soul.renderEmailSoulPromptFoundation();
The shared facade exposes name, description, whenToUse, whenNotToUse,
inputSchema, readOnly, and sideEffects. Handlers take a context containing
read, write, and network surfaces and the existing tool input. They return
{ ok, outcome, summary, next, data }; data.entities contains the message or
mailbox references, and next contains follow-up hints. Errors map to ok: false.
Read operations may refresh the local cache, so metadata conservatively marks
all tools as writing state.
const definitions = tools.aligned.listEmailToolDefinitions();
const tool = tools.aligned.getEmailToolByName('mail_read');
if (!tool) throw new Error('mail_read is unavailable');
const result = await tool.handler(
{ read: mailbox.read, write: mailbox.write, network: mailbox.network },
{ view: 'folders' },
);
console.log(result.ok, result.summary);
The deprecated tools.emailTools and tools.getEmailToolByName() remain
available with the original { outcome, summary, entities, nextSteps } result
and unchanged inputs. Migrate to tools.aligned to use the shared shape.
See LLM.md for the full AI-facing guide covering soul, tools,
and skills.
CLI (bash)
The package ships as a standalone email binary. Accounts are stored in
~/.config/email/accounts.json and every command accepts --json for
machine-readable output.
email account add --name work \
--imap-host imap.example.com --smtp-host smtp.example.com \
--user you@example.com --pass secret
email sync
email read
email search "invoice"
email compose send --to bob@example.com --subject "Hi" --body "Hello"
email read queue --json | jq '.entities[].title'
echo "body text" | email compose send --to bob@example.com --subject "Hi"
All commands return the same { outcome, summary, entities, nextSteps }
shape as the tool layer, with consistent exit codes (0 success, 1 user error,
2 config, 3 auth, 4 network, 5 tool error).
See SKILL.md for the complete CLI reference for agent operators
and Claude Code / OpenClaw users.
Tools
Five tools shaped around operator intent, not raw protocol operations:
mail_read | Read folder lists, message queues, threads, bodies, attachments, or raw EML |
mail_search | FTS5 local search or IMAP SEARCH on the server |
mail_compose | Send, reply, forward, save/update/send drafts |
mail_organize | Flags, labels, copy, move, archive, trash, junk, folder management |
mail_sync | Connect, disconnect, reconnect, sync, refresh folders, watch |
Key Properties
- Zero runtime dependencies. Only
node:sqlite, node:tls, node:net,
node:zlib, node:crypto (built into Node 24+).
- Single prebundled blob. One ESM + one CJS entry in
dist/. No subpath
exports, no code splitting.
- Pure SQLite storage. FTS5 full-text search, six tables, and deterministic
derived reads.
Mailbox creates and manages its own DatabaseSync instance.
- Full IMAP engine. Streaming tokenizer, response parser, COMPRESS=DEFLATE,
IDLE (RFC 2177), CONDSTORE/QRESYNC (RFC 7162), XOAUTH2/OAUTHBEARER.
- Full SMTP client. TLS/STARTTLS/plaintext, PLAIN/LOGIN/XOAUTH2, dot
stuffing, APPEND to Sent after send.
- Fetch-on-demand bodies. Initial sync pulls headers and BODYSTRUCTURE only.
Message bodies and attachment bytes are fetched when actually read.
- JWZ threading. Conversation trees computed from References/In-Reply-To
headers, with depth-ordered rendering and cross-folder support.
- Intention-shaped writes.
send, reply, forward, archive, trash,
saveDraft, sendDraft — operations that say what happened, not generic CRUD.
- Additive AI runtime.
soul for posture, tools for actions, skills for
workflow guidance — all optional, all structured.
- Colocated tests. Every non-type module has a colocated
.test.ts file.
The documented behaviour is backed by executable coverage.
Package Surface
import {
Mailbox,
initSchema,
read,
write,
network,
tools,
skills,
soul,
} from '@ghostpaw/email';
All domain and runtime types are also available at the root for TypeScript
consumers:
import type {
EmailConfig,
AuthConfig,
EmailReadSurface,
EmailWriteSurface,
EmailNetworkSurface,
EmailToolDefinition,
EmailToolResult,
EmailSkill,
EmailSoul,
ComposeInput,
ReplyInput,
ForwardInput,
Message,
MessageDetail,
Thread,
Folder,
Attachment,
AttachmentMeta,
SyncResult,
WatchEvent,
} from '@ghostpaw/email';
Documentation
| SKILL.md | Agent operators and Claude Code users running email as a bash tool |
| HUMAN.md | Human developers using the Mailbox API with read / write / network |
| LLM.md | Agent builders wiring soul, tools, and skills into LLM systems |
| docs/README.md | Architecture overview: model, invariants, protocol boundaries, source layout |
| docs/entities/ | Per-entity manuals with schema, operations, and invariants |
Development
npm install
npm test
npm run typecheck
npm run lint
npm run build
The repo is pinned to Node 24.14.0 via .nvmrc / .node-version /
.tool-versions / mise.toml / Volta. Use whichever version manager you
prefer.
Support
If this package helps your project, consider sponsoring its maintenance:

Anonyfox • MIT License