OpenCode plugin that acts as a sentinel between the AI agent and MCP servers — polling long-running tasks so token-costly status loops never enter the LLM inference path
View on GitHub ↗A plugin for OpenCode that acts as a sentinel between the AI agent and MCP servers — polling long-running tasks on the agent's behalf so that token-costly status loops never enter the LLM inference path.
When an agent submits a long-running job through an MCP tool, it must repeatedly call the server to check progress — each round-trip burns context window tokens.
opencode-mcp-sentinel moves the polling loop out of the agent and into the plugin runtime — 2 inference calls regardless of task duration.
Add to your opencode.jsonc (project-level .opencode/opencode.jsonc or global ~/.config/opencode/opencode.jsonc):
The plugin reads your existing MCP server configs — no additional setup needed.
Environment variables for controlling memory usage:
Variable Default Description ----------------------- --------- ------------------------------------------------- SENTINELMAXPOLLLOG unlimited Max poll log entries per task (FIFO trim) SENTINELTASKTTLMS unlimited Auto-cleanup completed tasks after N milliseconds
Both accept positive integers only. Zero, negative, or non-numeric values are treated as unlimited/disabled.
Submit a long-running MCP tool call and poll it at regular intervals until a condition is met. The sentinel polls silently (zero token cost) and notifies you when done.
Parameter Type Default Description ---------- ------ ---------- ------------------------------------------ server string required MCP server name (from opencode config) tool string required Tool name to call on the server args string "{}" JSON string of arguments for the tool interval number 5000 Pol
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/GCS-ZHN--opencode-mcp-sentinel.html)
runs anywhere. uses anything
pueschel88/Tradingview-MCPTradingview MCP server for controlling Tradingview Desktop from Claude Code via Chrome DevTools Protocol
0xsline/OpenChatCutLocal-first conversational AI video editor with a professional multi-track timeline, Agent Skills, MCP integration, and Remotion-powered rendering.
EndymionLee/PilotBrowseMCPA browser runtime that lets AI agents control your real Chrome browser via MCP. Agents can explore websites, generate operation manuals, and reuse them to save tokens. AI操控你的真实浏览器。
herry2059/project-os-for-codexOpen-source control plane for Codex projects: Git-backed context, visible agent progress, scoped MCP access, resumable work, and safe handoffs.
JesusRS1/stock-trade-finance-apiStock Trade MCP server in Tiingo with stock prices, news, forex, and fundamentals programmatically via the MCP
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.