addy-mcp

An MCP server that lets Claude (and other MCP clients) manage your addy.io account: aliases, recipients, usernames, custom domains, labels, rules, the blocklist and failed deliveries. It covers the full addy.io API.
Unofficial. This project is not affiliated with or endorsed by addy.io.
It runs on your own computer with your own API key. Nothing is hosted, and your key is only ever sent to addy.io.
1. Create an API key
In addy.io go to Settings โ API Keys (https://app.addy.io/settings/api) and click Create New API Key. Consider giving it a name like "Claude" and an expiry date.
2. Install
Claude Desktop (easiest)
- Download the latest
addy-mcp-<version>.mcpb from the Releases page.
- Double-click it (or drag it into Settings โ Extensions in Claude Desktop) and click Install.
- Paste your API key when asked. Claude Desktop stores it encrypted.
Claude Desktop (manual config)
Requires uv. Open Settings โ Developer โ Edit Config and add:
{
"mcpServers": {
"addy": {
"command": "uvx",
"args": ["addy-mcp@latest"],
"env": {
"ADDY_API_KEY": "paste-your-api-key-here"
}
}
}
}
Then fully quit and reopen Claude Desktop. If it can't find uvx, replace "uvx" with the full path from which uvx (macOS/Linux) or where uvx (Windows). Desktop apps don't always see your shell's PATH.
Claude Code
claude mcp add addy --scope user --env ADDY_API_KEY=paste-your-api-key-here -- uvx addy-mcp@latest
This makes the server available in Claude Code (the CLI and the desktop app's Code tab). Claude Desktop's chat uses a separate config, set up above.
Other MCP clients
Run uvx addy-mcp@latest as a stdio server with ADDY_API_KEY set in its environment.
@latest makes uvx check PyPI for a new release each time the server starts. Leave it off and uvx keeps using whichever version it first downloaded. To stay on a specific version instead, use e.g. addy-mcp==0.1.1.
Configuration
| Environment variable | Required | Description |
|---|
ADDY_API_KEY | Yes* | Your addy.io API key. |
ADDY_API_KEY_FILE | No | Path to a file containing the key, instead of ADDY_API_KEY. Takes precedence if both are set. *One of the two is required. |
ADDY_READ_ONLY | No | Set to true to hide every tool that can change your account. Claude can then only look things up. |
ADDY_BASE_URL | No | For self-hosted addy.io instances, e.g. https://addy.example.com. Defaults to https://app.addy.io. |
Storing your key securely
The setup above keeps the key in plain text in your MCP client's config file. That's the usual approach for local MCP servers, but you can do better:
-
Claude Desktop extension: the .mcpb install stores the key encrypted. Nothing else to do.
-
macOS Keychain (Claude Code and other clients): keep the key in the Keychain and fetch it only when the server starts, so it's never written to a config file. Add the key (you'll be prompted for it, so it doesn't end up in your shell history):
security add-generic-password -a "$USER" -s addy-mcp -w
Then add the server with a small wrapper instead of --env:
claude mcp add addy --scope user -- /bin/sh -c 'ADDY_API_KEY="$(security find-generic-password -a "$USER" -s addy-mcp -w)" || exit 1; export ADDY_API_KEY; exec uvx addy-mcp@latest'
If the server fails to start because uvx isn't found, replace uvx with its full path from which uvx.
On Linux, the same pattern works with secret-tool store --label=addy-mcp service addy-mcp and secret-tool lookup service addy-mcp. With 1Password, use op read "op://<vault>/<item>/credential" in place of the security command.
-
Key file: set ADDY_API_KEY_FILE to a file that only you can read (chmod 600), or to a Docker or Kubernetes secret mount.
-
VS Code: let VS Code prompt for the key and keep it in its secret storage:
{
"inputs": [
{ "type": "promptString", "id": "addy-api-key", "description": "addy.io API key", "password": true }
],
"servers": {
"addy": {
"type": "stdio",
"command": "uvx",
"args": ["addy-mcp@latest"],
"env": { "ADDY_API_KEY": "${input:addy-api-key}" }
}
}
}
Whichever you choose, use a separate addy.io key for each app or device, so you can revoke one without breaking the others, and give it an expiry date.
Updating
- Installed with
addy-mcp@latest (the commands above): restart Claude (or your MCP client) and the newest release is picked up automatically.
- Installed with plain
uvx addy-mcp: uvx keeps using the version it first downloaded. Change addy-mcp to addy-mcp@latest in your config and restart Claude.
- Claude Desktop extension (
.mcpb): download the new .mcpb from the Releases page and install it the same way. Your API key and settings are kept, so you won't be asked for them again.
To see which version is newest, run uvx addy-mcp@latest --version (available from 0.1.1). To hear about new releases, watch the repo on GitHub (Watch โ Custom โ Releases). Changes are listed in the changelog.
What you can ask
- "Make me a new alias for my Netflix account."
- "Which of my aliases have blocked the most emails? Deactivate the top 5."
- "Find every alias I created for shopping sites and label them 'Shopping'."
- "Block everything from marketing.example.com."
- "Create a rule that blocks newsletters (List-Unsubscribe header) sent to my shop@ alias."
- "Why did mail to my bank alias fail last week?"
| Area | Tools |
|---|
| Account | get_account_details, get_api_token_details, get_account_notifications, get_domain_options, get_app_version |
| Aliases | list_aliases, get_alias, create_alias, update_alias, delete_alias, restore_alias, forget_alias, set_alias_option (active / pinned / attached-recipients-only), set_alias_recipients, set_alias_labels |
| Bulk aliases | bulk_alias_action (get / activate / deactivate / pin / unpin / delete / restore / forget), bulk_set_alias_recipients, bulk_set_alias_labels |
| Recipients | list_recipients, get_recipient, create_recipient, update_recipient, delete_recipient, resend_recipient_verification, add_recipient_public_key, remove_recipient_public_key, set_recipient_option (active / can reply-send / encryption / inline encryption / protected headers / strip PGP keys / strip PGP signatures) |
| Usernames | list_usernames, get_username, create_username, update_username, delete_username, set_username_default_recipient, set_username_option (active / catch-all / can login) |
| Domains | list_domains, get_domain, create_domain, update_domain, delete_domain, set_domain_default_recipient, check_domain_sending_records, set_domain_option (active / catch-all / shared with family) |
| Labels | list_labels, get_label, create_label, update_label, delete_label |
| Rules | list_rules, get_rule, create_rule, update_rule, delete_rule, reorder_rules, set_rule_active |
| Blocklist | list_blocklist, add_to_blocklist, remove_from_blocklist |
| Failed deliveries | list_failed_deliveries, get_failed_delivery, download_failed_delivery, resend_failed_delivery, delete_failed_deliveries |
Every tool is labelled read-only, write, or destructive (MCP tool annotations), so clients like Claude can ask before running anything risky. Deleting an alias can be undone with restore_alias. Forgetting an alias is permanent.
Privacy and security
- The server runs locally and talks only to addy.io (or your
ADDY_BASE_URL). It has no telemetry.
- Your API key gives full access to your addy.io account. Use a dedicated key so you can revoke it on its own, and turn on
ADDY_READ_ONLY if you only want Claude to look things up.
- Anything the tools return (alias addresses, descriptions, failed-delivery emails) is passed to your AI client, and is subject to that client's privacy policy.
Development
git clone https://github.com/Former-Stranger/addy-mcp
cd addy-mcp
uv run --no-editable python -m pytest
uv run ruff check && uv run ruff format --check
scripts/live_test.py exercises the server against a real account: it creates, updates, deletes, restores and finally forgets a throwaway alias and label. Run it with ADDY_API_KEY=... uv run --no-editable python scripts/live_test.py --yes.
--no-editable: on some macOS setups files in .venv get the "hidden" flag, and Python 3.13+ ignores hidden .pth files, which breaks editable installs. A regular install sidesteps that.
Releasing
- Bump
version in pyproject.toml, run uv run python scripts/sync_metadata.py, and update CHANGELOG.md.
- Commit, then tag and push:
git tag v0.2.0 && git push origin v0.2.0.
- Approve the
pypi deployment when GitHub asks (the release pauses until a maintainer approves it).
The release workflow publishes to PyPI, attaches the .mcpb bundle to a GitHub release, and updates the MCP Registry listing.
A weekly workflow compares the addy.io API docs with scripts/addy_api_snapshot.txt and opens an issue when the API changes. After updating the tools, refresh the snapshot with python3 scripts/check_api_docs.py --update. Dependabot opens weekly pull requests for dependency and GitHub Actions updates.
License
MIT