Enables local AI agents to communicate directly over Unix sockets, allowing multi-agent collaboration across different models and harnesses without any broker or network dependency.
Enables authenticated clients to discover and call tools for scoped file listing/searching, request client-side LLM sampling, and receive real-time progress/log notifications over stdio or secure Streamable HTTP, with tenant/user workspace isolation and a loopback browser dashboard.
Exposes a context_folding MCP tool that lets an agent mark conversation history entries (user messages, model replies, tool results) as foldable, keep, or unmarked and replace marked content in outgoing model requests with a head, a concise summary, and an original-entry ID, without altering the client's own full history. Enables listing, paginated reading of originals, layered expansion via previous IDs down to the original, status inspection, and cleanup, with folding applied only when token thresholds and trusted cost-accounting gates are satisfied.
Enables compatible agents to call registered local tools and read resources through one stdio server, including harness preparation, knowledge search and document management, memory retention/recall, correction proposals, advisory decisions, tool/skill discovery, and log/source inspection.
Exposes registered AI agents as MCP tools over stdio, enabling clients to invoke Git operations, agent-building assistance, bookmark management, and project-idea generation through OpenAI-backed interpreters.
Enables an AI agent to run local tools on the user's own machine via stdio, including command execution, workspace file read/write, and system status checks.