Skip to content

tokencanopy/e2a

v0.9.4Apache-2.0

Open-source email API for applications and AI agents — transactional sending from any product, per-agent two-way inboxes, structured SPF/DKIM/DMARC evidence, and a queryable event log. 78 MCP tools over hosted streamable HTTP with OAuth.

GitHubOpen Website

e2a plugin

The open-source email API for applications and AI agents. Send transactional email from any product, give agents real two-way inboxes, and keep people in control. Installing this plugin registers the hosted e2a MCP server (https://api.e2a.dev/mcp, Streamable HTTP + OAuth 2.1) and five stable skills for setup, application integration, diagnosis, evaluation, and everyday inbox operation. Installing core supplies the MCP connection used by both core and the optional Labs workflows.

After installation, authorize e2a through your client's MCP flow (Claude Code: run /mcp; Codex CLI: codex mcp login e2a) — no API key to paste. For headless/CI, an account API key works too; see clients/.

With an e2a inbox, developers can build support agents, AI receptionists, scheduling agents, e-commerce agents, sales agents, recruiting agents, voice follow-up agents, and procurement agents that receive real email, keep conversation state, and reply safely. See the complete use-case catalog.

Install

The same plugin ships native manifests for Claude Code, Codex, and Cursor.

Claude Code

claude plugin marketplace add tokencanopy/e2a
claude plugin install e2a@e2a

Codex

codex plugin marketplace add tokencanopy/e2a

Then launch codex, run /plugins, search for e2a, and install — it walks you through the OAuth path. (Codex desktop: Plugins → Add more + → paste https://github.com/tokencanopy/e2a.)

Cursor

Point Cursor at the hosted MCP server. Project-level config lives in .cursor/mcp.json, global in ~/.cursor/mcp.json (project wins on conflict):

{
  "mcpServers": {
    "e2a": { "url": "https://api.e2a.dev/mcp" }
  }
}

Remote servers take url only — type/transport are stdio-only in Cursor. On first use Cursor registers itself via OAuth Dynamic Client Registration and opens your browser; there is no API key to paste and no auth block to fill in.

Cursor lists core and receives its MCP configuration and canonical setup documentation. Labs is not listed, so Labs skills are unavailable in Cursor. Install Labs only in Claude Code or Codex.

This file used to recommend two things that don't work, so they're worth naming before someone re-adds them: a bare /add-plugin e2a resolves only against Cursor's curated marketplace, which e2a isn't published to; and there is no "paste a repo URL into marketplace search" flow — importing a repo is Dashboard → Plugins → Add Marketplace → Import from Repo, which creates a team marketplace and is Teams/Enterprise only.

Other MCP clients (manual)

Clients without native plugin support (Zed, Goose, Windsurf, Claude Desktop, raw mcp.json) can point straight at the hosted server. Ready-to-paste configs are in clients/; the full per-client guide is at https://e2a.dev/setup.md.

What's inside

plugins/e2a/
├── plugin.meta.json             # SOURCE OF TRUTH — every manifest below is generated from it
├── plugin.json                  # Agent Plugins v1 portable manifest (generated)
├── mcp.json                     # Agent Plugins v1 portable MCP config (generated)
├── .claude-plugin/plugin.json   # Claude Code manifest (generated)
├── .codex-plugin/plugin.json    # Codex manifest, skills + mcpServers + interface (generated)
├── .cursor-plugin/plugin.json   # Cursor manifest (generated)
├── .mcp.json                    # the hosted MCP server, legacy clients (generated)
├── assets/icon.svg
├── docs/                        # canonical agent docs mirrored at e2a.dev
│   ├── setup.md                 # connect guide + first-inbox workflow
│   ├── auth.md                  # OAuth, API keys, scopes, agent identity
│   ├── sdk.md                   # SDK + webhook integration guide
│   ├── templates.md             # email-template guide
│   └── llms.txt                 # machine-readable hosted docs index
├── skills/                      # five stable skills
│   ├── e2a/SKILL.md             # everyday inbox operation
│   ├── e2a-setup/SKILL.md       # MCP, OAuth, inbox, and domain readiness
│   ├── e2a-integrate/SKILL.md   # SDK or REST application integration
│   ├── e2a-doctor/SKILL.md      # evidence-backed diagnosis and repair
│   └── email-evals/SKILL.md     # deterministic synthetic email-agent evaluation (/email-evals)
└── clients/                     # manual paste-in configs for non-plugin clients

/email-evals guides synthetic, deterministic email-agent cases through contained validation, explicit approval, and report-driven iteration.

Experimental autonomous workflows (/agentify, /autopilot, and /tether) live in the separate e2a-labs package. Install core e2a first: it remains the sole owner of the hosted MCP connection.

The marketplace manifests that expose this plugin live at the repo root: .claude-plugin/marketplace.json, .cursor-plugin/marketplace.json, and .agents/plugins/marketplace.json (Codex).

Developing

Skills are authored in skills/<name>/SKILL.md with YAML frontmatter:

---
name: e2a
description: Use when operating e2a over its MCP tools — sending/receiving email, ...
version: 12
---

...guide body...
  • name (required) — must match the directory; lowercase letters, digits, hyphens; ≤64 chars.
  • description (required) — write it as "Use when…"; this is how Claude Code decides to load the skill. ≤1024 chars.

node scripts/validate-plugin.mjs (run by the Plugin tests / Package and manifests CI job) validates both plugin packages independently: their exact client manifest sets, per-package manifest versions, marketplace source paths, sole core ownership of the MCP server, and every SKILL.md frontmatter. It also regenerates every manifest and fails if one drifted from the per-package plugin.meta.json, checks each package's Agent Plugins v1 portable manifest against the closed v1 schemas, and rejects any skill file interpolating a client-specific path variable such as ${CLAUDE_PLUGIN_ROOT} (those expand on one client and land as literal text everywhere else). node --test scripts/validate-plugin.test.mjs covers those guards against fixture trees. A change that wouldn't load fails CI.

Agent-facing Markdown is authored in docs/. Run node ../../scripts/sync-agent-docs.mjs from this directory (or node scripts/sync-agent-docs.mjs from the repository root) to refresh its committed web/public/ mirrors. The repository-integrity CI job runs the same script with --check and fails if a hosted mirror drifts from its canonical source.

Never hand-edit a manifest. The portable pair, the client manifests, .mcp.json, and the three root marketplace.json files are all generated. Edit the package's plugin.meta.json (plugins/e2a/plugin.meta.json or plugins/e2a-labs/plugin.meta.json), then run:

node scripts/generate-plugin-manifests.mjs

and commit the source together with whatever it regenerated. Editing a generated file directly fails CI, naming the file to fix. That includes version bumps: change version in the package's plugin.meta.json and regenerate. Marketplace metadata always follows the core version; Labs is independently versioned.

Reference