wechat-mcp-manual

# WeChat MCP Use the schema for routine calls; this progressive disclosure guide adds media, setup/recovery and unfamiliar result details, not a per-send reading ritual. ## First safe action ```json {"action":"check","input":{},"reasoning":"find WeChat conversations"} ``` Without a complete current notification, recover the needed content using `read` with an exact `user_id` from `check`, `read`, `contacts`, or a current notification. For `reply`, use an inbound `message_id` from `read`, `search`, or a current notification's structured preview (`recent_messages`/ `latest_incoming`); never guess or invent one. When a CURRENT notification carries the complete required message and exact `id`, do not call `check`/`read` merely to reread that identical content or to re-obtain an `id` already present there. `search` matches inbox bodies by regex; it cannot prove that an outgoing reply is absent. Calls require `action`, closed action-owned `input`, and `reasoning`; optional `summarize` is boolean. The advertised schema owns the action catalog and validation. ## External effects and recovery `send`/`reply` are real external side effects: obtain authority and verify recipient and content. Success is provider acceptance, not delivery (`delivery_confirmed` is false); never replay an accepted request. `reply` cannot fall back to a new recipient if its inbound target or sender is missing. `media_path` is a file inside the allowed working directory. Text precedes media; partial results or uncertain errors require reconciliation, not whole-request replay. A missing local sent record does not prove nothing was accepted. Read [operations](reference/operations.md) for result/read-state limits and [media](reference/media.md) for paths, validation and upload stages. `getUpdates` has one consumer per bot account. The per-account lock refuses a second poller; do not delete its lockfile or start another. A refresh, worker failure, or recovery event is a lifecycle fact, not by itself a reason to reread a complete current notification; call `read` to reconcile only when the inbox/sent history you actually need is missing or uncertain after such an event, not as a routine step before every reply. Local history is context, not a delivery receipt. Login, configuration and credential replacement require the owner's setup authorization. Avatar sessions must not reconfigure this MCP. Never share the admin login QR, token or credentials. See [setup/recovery](reference/setup.md); host addon activation remains owned by `mcp-manual`. ## Settings SHOW `settings` takes `input={}` and returns the startup snapshot without writing or rereading owner files. Five sensitive rows redact both `current` and `default`; only `poll_interval` is displayed. An unavailable row fails the whole inventory. SHOW grants no configuration authority. Each stable anchor routes directly: ## setting-config-path [Resolved config source](reference/setup.md#setting-config-path). ## setting-base-url [Endpoint precedence](reference/setup.md#setting-base-url). ## setting-poll-interval [Polling interval/default](reference/setup.md#setting-poll-interval). ## setting-allowed-users [Inbound allow-list](reference/setup.md#setting-allowed-users). ## setting-bot-token [Login secret](reference/setup.md#setting-bot-token). ## setting-user-id [Backend identity](reference/setup.md#setting-user-id). Sidecars ship with the package but are not embedded in `manual`; follow the relative links only when their detail is needed.