MCP server that lets any MCP host (Claude Code, Codex CLI, Cursor, etc.) drive OpenCode and delegate tasks asynchronously to its agents
OpenCode MCP has 9 well-structured tools with consistent naming (verb_noun pattern: opencode_start_server, opencode_list_agents, etc.). All tools have descriptions (avg ~80 chars) and Zod-validated input schemas with typed parameters. However, descriptions are terse and lack context about when to use each tool or what happens on failure. Output schemas are not documented, callers must infer response structure from code. Error handling returns JSON but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels (WRITE, READ_ONLY, REVERSIBLE). Parameter descriptions are minimal; e.g., 'Id of the server instance to query' doesn't explain what data is returned or when to call this vs. another tool.
Cancel a delegated task by aborting its OpenCode session
Send a follow-up prompt to an existing delegated task's session
Get the final result of a completed task
Get the current status of a delegated task
List available agents and models for a running OpenCode server
Start a headless OpenCode server instance in the current working directory
Start a delegated task on an OpenCode server
Output schemas not documented. Callers cannot infer response structure (e.g., what fields does opencode_start_server return? what does opencode_list_agents return?). LLMs must guess or reverse-engineer from code.
Tool annotations missing. Tools have explicit risk levels (WRITE, READ_ONLY, REVERSIBLE) in metadata but do not expose readOnlyHint, destructiveHint, or idempotentHint in schema. Agents cannot determine safety without reading descriptions.
Error handling lacks recovery guidance. Errors return JSON with status/message but do not suggest next steps. E.g., 'server_id not found' should suggest 'Call opencode_start_server() to create one' or 'List running servers with opencode_list_agents()'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Stop a running OpenCode server instance
Wait for one or more delegated tasks to complete
Descriptions are terse and lack context. 'List available agents and models for a running OpenCode server' does not explain what agents/models are, when to call this vs. opencode_start_task, or what the response structure is.
Parameter descriptions lack format/constraint details. E.g., 'agent' param in opencode_start_task says 'Agent name to delegate to (e.g. build, plan)' but does not state: are these the only valid values? Is it case-sensitive? What happens if omitted?