Use this to delegate a new coding task to OpenCode and run its first turn. Blocks until the turn finishes (or wait-seconds elapses, up to the turn timeout). The final answer and status are in structuredContent.content / structuredContent.status; the text response mirrors the same content. Returns a session id (also usable as threadId/conversationId) to continue with opencode-reply. sandbox selects a cooperative permission profile for OpenCode's own tools (read-only / workspace-write / danger-full-access) — it is not OS-level isolation, OpenCode can still run arbitrary shell commands unless a rule denies it. approval-policy controls permission prompts raised by OpenCode: never (default) auto-denies them, on-request asks interactively when the client supports it. Cancelling or stopping the call that OWNS this turn (Esc, TaskStop, a client timeout) cancels the OpenCode turn; a call that only joined an existing turn as an observer (e.g. a duplicate request-id) detaches without cancelling anything. To stop waiting but keep the turn running, pass wait-seconds and poll opencode-status. Use detail/max-output-chars to keep this response small (the full answer stays readable with opencode-output); request-id makes a retried call safe to repeat; output-schema asks the final message to be one JSON value matching a schema. To run several tasks in parallel, start each with wait-seconds:0 and check them together with opencode-status ids/wait-for.

Use this to continue an existing OpenCode session started with opencode, after reading its previous result from structuredContent.content. Requires exactly one of sessionId, threadId or conversationId, plus a new prompt. Blocks until the turn finishes (or wait-seconds elapses, up to the turn timeout); the answer is in structuredContent.content. sandbox and approval-policy stay fixed for the life of the session. Cancelling or stopping the call that OWNS this turn (Esc, TaskStop, a client timeout) cancels the OpenCode turn; a call that only joined an existing turn as an observer (e.g. a duplicate request-id) detaches without cancelling anything. To stop waiting but keep the turn running, pass wait-seconds and poll opencode-status. Same detail/max-output-chars/request-id/output-schema options as opencode; output-schema applies only to this turn, never a stored default.

Use this to check on or wait for a running OpenCode turn, or — with no id — to list the sessions tracked by this server. With an id it can block up to wait-seconds (max 600) and will also relay any pending approval prompts to you, so it is not purely read-only: it can act on the session while waiting. Without an id it returns the tracked session list in structuredContent.sessions and structuredContent.content. Pass ids (up to 16, instead of a single sessionId/threadId/conversationId) to check several sessions in one call — pair with wait-seconds:0 on opencode/opencode-reply to fan work out in parallel, then wait-for "any" (default) or "all" here to collect the results; batch items are always compact and max-output-chars becomes the shared answer budget split across them. detail/max-output-chars shrink a single-session answer the same way as on opencode.

Use this to stop the turn currently running on an OpenCode session without ending the session, so a new opencode-reply can be sent afterwards. Requires exactly one of sessionId, threadId or conversationId. Idempotent when the session is already idle; the result is in structuredContent.content.

Page through a turn's full retained answer, tool-call log, structured output, or the OpenCode-reported file diff for that turn. Use detail:"compact" and max-output-chars on opencode/opencode-reply/opencode-status to keep every normal answer small; come back here with the same turn number to read more of it, one page at a time (offset/limit), without paying for the full text on every call — useful together with wait-seconds:0 plus opencode-status ids/wait-for fan-out, reading each turn's full output here only for the ones that need it. Requires exactly one of sessionId, threadId or conversationId, plus turn; section defaults to "answer". Retained output expires (OUTPUT_UNAVAILABLE) and is dropped once opencode-end succeeds, so read what you need before ending the session. Read-only: never mutates OpenCode.

Discover this server's configuration and OpenCode's current catalogs without starting a session. section "server" (default) reports this server's version, mode, defaults and limits (maxOutputChars, maxBatchIds, output retention, request-id table size, …); "models" and "agents" list what OpenCode currently advertises for a given cwd; "roots" lists this server's allowed working-directory roots. "models"/"agents"/"roots" are paginated (offset/limit) and return a snapshot-id to keep reading the same consistent snapshot; pagination arguments are invalid for "server". Read-only; the "server" section never starts OpenCode.

Use this to finish working with an OpenCode session: it stops any running turn, rejects any leftover permission prompts, then deletes or archives the session and forgets it (action, default 'delete' — this server's configured OPENCODE_MCP_END_ACTION). Requires exactly one of sessionId, threadId or conversationId. Idempotent: ending an unknown or already-ended session returns structuredContent.status "not_found" without contacting OpenCode.