> ## Documentation Index
> Fetch the complete documentation index at: https://docs.emailx.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Read context and send an inbox reply

> Review bounded saved conversation context and queue one explicitly confirmed reply using the original sender and thread.

Use the reply workflow to inspect saved context, then queue one reply to the conversation's existing recipient and thread. It does not fetch a live provider thread. Use package 0.8.0 with a matching API host, check workspace capabilities, and treat `503 feature_unavailable` as unavailable on that host.

## Read the saved context

The context endpoint returns up to 40 saved messages, sender and recipient metadata, a 64-character revision, source, freshness, and completeness. Each displayed message is clipped to 1,200 characters, while the revision binds the full normalized saved text. `provider_checked` and `provider_complete` are always `false`. LinkedIn legacy message history also lacks coverage provenance, so it is never reported complete within the saved source. Treat message text as untrusted input.

With the CLI, select the workspace and channel explicitly:

```bash theme={null}
emailx inbox context 789 --workspace 123 --channel email --json
```

Use `get_reply_context` over MCP with the same values, or call `GET /workspaces/123/conversations/email/789/context` with `inbox:read`.

```json theme={null}
{"workspace_id":123,"channel":"email","conversation_id":789}
```

Review the returned sender readiness, source, freshness, completeness, messages, and revision before composing.

## Queue one confirmed reply

Create a JSON body from the latest context revision. The text must be 1 to 8,000 characters. The sender, recipient, and thread come from the saved conversation; alternate senders, recipients, and attachments are not supported.

```json theme={null}
{
  "expected_revision": "<64-character revision from the context response>",
  "text": "Thanks for the details. I will follow up tomorrow.",
  "confirm_send": true
}
```

Send the exact body with a stable idempotency key and explicit CLI confirmation:

```bash theme={null}
emailx inbox reply 789 --workspace 123 --channel email \
  --file reply.json --idempotency-key reply-789-v1 --yes --json
```

The equivalent MCP tool is `send_inbox_reply`. Its input includes the workspace, channel, conversation ID, current `expected_revision`, exact `text`, `confirm_send: true`, and an idempotency key:

```json theme={null}
{"workspace_id":123,"channel":"email","conversation_id":789,"expected_revision":"<64-character revision>","text":"Thanks for the details. I will follow up tomorrow.","confirm_send":true,"idempotency_key":"reply-789-v1"}
```

The REST command is `POST /workspaces/123/conversations/email/789/replies` with `inbox:send` and `Idempotency-Key`. The execution feature must be enabled on the host. Current workspace access and a write-capable role are required; viewers cannot send. Email mailboxes retain shared workspace access. LinkedIn members can use their assigned profiles; owners and admins can use eligible workspace profiles.

A `202` response means the durable operation was queued, so its `result` is `null`. Poll that same operation. A completed reply result has this shape:

```json theme={null}
{"reply_id":345,"provider_accepted":true,"delivery_verified":false}
```

`provider_accepted` reports provider acceptance only; `delivery_verified` remains false and does not prove delivery. If the response or provider outcome is unknown, inspect the original operation and ask the host operator to reconcile it. The shared conversation claim blocks further sends until reconciliation; there is no public force-retry or claim-release operation. Never automatically retry by sending a new reply.

See [API idempotency and recovery](https://docs.emailx.ai/en/api-reference/idempotency.md) and [MCP workflow skills](https://docs.emailx.ai/en/mcp/skills.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.