tokencanopy · jiashuoz · Aug 17, 2026 · Aug 17, 2026
@@ -319,6 +319,7 @@ domains/webhooks in the **web dashboard**.
 | `e2a suppressions list\|add\|remove` | Inspect and manage recipient block lists (account-wide GA; agent-scoped is beta) |
 | `e2a send` / `e2a reply` | Send an email as the agent, or reply in-thread |
 | `e2a messages list\|get` | List or fetch messages for an agent |
+| `e2a metrics [<email>]` | Print delivery counters — account rollup, or one inbox (`--by-agent`, `--by-day`, `--json`) |
 | `e2a listen --agent <email>` | Stream inbound email for an agent over WebSocket (real-time; `--json` for raw, `--forward <url>` to bridge to a local HTTP handler) |
 | `e2a config [list\|get\|set]` | View or update the local config |
 

@@ -257,6 +257,25 @@ a server-owned, read-only mailbox-local identity. Human-readable formats do
 not change, and there is no `threadId` request flag, filter, or thread
 endpoint.
 
+### `e2a metrics`
+
+Print delivery counters for the account, or a single inbox.
+
+```bash
+e2a metrics                                          # account rollup
+e2a metrics bot@acme.com                              # one inbox
+e2a metrics --by-agent --json                         # account rollup broken down per agent
+e2a metrics --by-day --start 2026-08-01T00:00:00Z --json
+```
+
+With no positional argument this prints the account rollup; with one, that
+inbox's counters (accepted/submitted/delivered, bounces by class, complaints,
+suppressions, inbound DMARC results, review outcomes, and webhook delivery
+health). Flags: `--start`/`--end` (RFC 3339 window bounds), `--by-agent`
+(break the account rollup down per agent — account rollup only), `--by-day`
+(UTC daily buckets — account rollup only), `--json` (print the full metrics
+view as JSON instead of the human-readable summary).
+
 ### `e2a contacts` (beta)
 
 Manage account-level contact identity and per-agent outreach state, with

@@ -116,7 +116,7 @@ design.
 
 - **Symptoms**: requests succeed, but the credential is served at
   least-privilege **agent scope**: account-scoped callers see the tool list
-  shrink from 76 tools to the 20 runtime tools, and the per-request default
+  shrink from 78 tools to the 21 runtime tools, and the per-request default
   agent is unset (explicit `email` required).
 - **Detection**: `auth_resolution` WARNING `whoami probe failed; serving
   least-privilege fallback`; `mcp_auth_resolutions_total{result="fallback"}`.

@@ -121,15 +121,15 @@ Hosts that support OAuth connectors can instead add `https://api.e2a.dev/mcp` as
 
 ## Tools
 
-The server exposes up to **76** tools spanning agents, messages, human-in-the-loop
+The server exposes up to **78** tools spanning agents, messages, human-in-the-loop
 approval, attachments, domains, events, webhooks, API keys, contacts/outreach,
 email templates (beta), and suppressions.
 **The visible set depends on your credential's scope:** an **agent**-scoped
-credential sees the 20 runtime/inbox tools (read, send, reply, restore
+credential sees the 21 runtime/inbox tools (read, send, reply, restore
 messages, per-agent outreach); an **account**-scoped credential also sees the
-56 admin/setup tools (agent/domain/webhook/event/template/API-key/contact/
+57 admin/setup tools (agent/domain/webhook/event/template/API-key/contact/
 suppression management — **and HITL review discovery plus approve/reject, which
-is an account-owner action, never agent self-approval**) — all 76.
+is an account-owner action, never agent self-approval**) — all 78.
 Every tool carries MCP annotations (`readOnlyHint`/`destructiveHint`/
 `idempotentHint`) so hosts can auto-approve reads and flag destructive actions.
 The tables below highlight the most commonly used ones — your MCP host's tool list

@@ -12,7 +12,7 @@ End-to-end demos showing how to wire the e2a MCP surface into popular agent fram
 
 ## One hosted endpoint, same tool surface
 
-Each example exercises the same e2a tool surface (76 tools; the visible set depends on your key's scope) against the hosted MCP server. The Python framework examples ship an `agent.py` script; the Codex example ships a TOML block (Codex is itself the agent, so you configure it instead of writing a script).
+Each example exercises the same e2a tool surface (78 tools; the visible set depends on your key's scope) against the hosted MCP server. The Python framework examples ship an `agent.py` script; the Codex example ships a TOML block (Codex is itself the agent, so you configure it instead of writing a script).
 
 Every example connects to `https://api.e2a.dev/mcp` over Streamable HTTP with an
 agent-scoped API key in the `Authorization` header. Set `E2A_MCP_URL` to point

@@ -2,7 +2,7 @@
 
 [Codex CLI](https://github.com/openai/codex) is OpenAI's terminal coding agent (peer to Claude Code). It speaks MCP natively — you declare servers in `~/.codex/config.toml` and Codex connects to them at session start, then surfaces their tools to whatever model you're running.
 
-Unlike the [LangChain](../langchain/), [CrewAI](../crewai/), [Google ADK](../adk/), and [OpenAI Agents SDK](../openai-agents/) examples — which are Python scripts you run — Codex is itself the agent. The "example" here is the TOML you paste into your Codex config to add e2a's MCP tools (76 total; the visible set depends on your key's scope).
+Unlike the [LangChain](../langchain/), [CrewAI](../crewai/), [Google ADK](../adk/), and [OpenAI Agents SDK](../openai-agents/) examples — which are Python scripts you run — Codex is itself the agent. The "example" here is the TOML you paste into your Codex config to add e2a's MCP tools (78 total; the visible set depends on your key's scope).
 
 Codex opens a Streamable HTTP session to the hosted endpoint at `https://api.e2a.dev/mcp` with your API key in the `Authorization: Bearer …` header — no Node toolchain on the host (works on CI runners, dev containers, remote app servers, etc.).