name: innernet title: innernet repo: https://github.com/your-innernet/innernet-mcp sha: 9da4f24e04dffc526129c11546eb4bb2ef60ef81 subdir: hermes version: "1.4.0" description: >- One memory for every AI tool: load a project's context, answer from what was decided before, save new decisions back, and pick up where the last session left off, in the same memory Claude, ChatGPT, Cursor and Grok read and write. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/) pointing at innernet's hosted MCP server (Streamable HTTP, OAuth 2.1 with CIMD and DCR); the two bundled skills load as innernet:innernet and innernet:innernet-handoff. Disclosure: hosted-only connector with no local code, hooks, background process, telemetry or self-update. Every tool call goes to innernet.live, signed in as the user through innernet's own browser sign-in (a first sign-in creates a free innernet account); what a call carries (a project name, a question, text the user asks to save) is sent there, and what the memory holds comes back into the session context, so it reaches the user's model provider. Some tools write to the user's innernet account (save context, capture a note, write a handoff, update a task, forget on request); every write is versioned and can be read, edited or deleted on the user's dashboard. The tool list is served by the remote server at run time and is not covered by the pinned sha. Hermes stores and refreshes the OAuth tokens. OAuth must be enabled with an `mcp_servers.innernet` entry in config.yaml (url https://innernet.live/api/mcp, `auth: oauth`) and `hermes mcp login innernet`; the bundled mcp.json alone yields 401. maintainer: your-innernet tier: community category: tools requires_hermes: ">=0.22.0" docs_url: https://innernet.live/docs/everywhere#hermes-agent readme: true platforms: [] capabilities: provides_tools: [] provides_hooks: [] provides_middleware: [] requires_env: []