What it does
The MCP server at getargentic.com/mcp connects a coding agent to Argentic, so it can wake agents that keep their memory between calls, read what they remember, give them tools, and manage API keys, usage and budget. The repository contains manifests, skills, MCP settings and small client adapters, but not the engine. It also includes crash-test, a small open tool that stops an agent at each tool call, restarts it and reports which actions ran twice, for agents built on any framework.
This suits anyone running agents over long periods who wants to see which runs were cut off, resume them, or check how an agent recovers from interruption. The adapters let developers keep tools already written for the Vercel AI SDK, Mastra, LangChain or LangGraph, Pydantic AI or the OpenAI Agents SDK and hand the loop to Argentic. A scripted comparison with a plain loop is also available.
The MCP needs no API key: you sign in with your Argentic account over OAuth and choose an organization. Waking an agent also requires your own model provider key, which is sent in the X-Model-Key header on each request and is not stored by Argentic. Setup steps are given for Claude Code, claude.ai, Cursor and Codex. The plugin adds five skills: argentic, argentic-setup, argentic-quickstart, argentic-status and argentic-fix.
Limitations and requirements
Requires an Argentic account and a sign-in via OAuth. Waking agents requires your own model provider key. Running your own agent or the check tool requires an Argentic API key and a model key. The argentic_answer_approval tool requires an owner or admin role; other tools are limited by your role. On claude.ai, the Request headers option is a beta that not every account has. The Python package is not yet on PyPI and must be installed from the repository until it is. The demo command in client 0.2.0 and later has been removed. The publisher states the project is MIT licensed.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
Argentic
Run agents for hours. Argentic keeps their context sharp and records every step.
v0.3.3As shown on registry.modelcontextprotocol.io when we last read it