Multi-agent department framework for long-form complex tasks, fighting AI hallucination, validated on academic research. 共识管线：多智能体部门长线任务解决框架，对抗AI幻觉，以学术研究为验证场景。

Human-in-the-loop (HITL) for AI agents: ask a person for approvals or answers by push notification. One command connects Claude Code, Codex, Cursor, Gemini CLI and more.

Local MCP server that gives coding agents a verifiable project memory, trims what enters their context, and masks secrets — with honest accounting of what it saves and costs.

Manage your team — create roles, assign tasks, spawn workers, and monitor progress

Exposes the full Plexim PLECS Standalone JSON-RPC interface as MCP tools, letting agents auto-start and connect to PLECS to load and edit models, run simulations and analyses, manipulate scopes and webserver data, and perform code generation or FMU export with structured, size-aware results. Includes write guardrails, concurrency handling for long simulations, and batch/raw passthrough escape hatches.

Human-in-the-loop (HITL) for AI agents: ask a person for approvals or answers by push notification. One command connects Claude Code, Codex, Cursor, Gemini CLI and more.

Iterative worker-reviewer cycle that spawns a critic subagent to score work 1-10 and provide actionable feedback, then revises until a quality gate is met. Use when implementing features, writing specs, reviewing existing code, or completing any task where quality matters more than speed. Trigger ph

Turns a Muse chat ID into an addressable endpoint so any tool-capable process can send messages, ask questions and read replies, follow live activity, and create, rename, archive, or delete chats over the official Muse gadget Link SDK.

Enables agents to query real-time GPU rental prices across 26 platforms, including current lowest quotes, 30-day price history, market movers, platform reliability, and task/budget recommendations. Access is read-only and metered per call via x402 with USDC on Base, requiring no signup or API key.

Enables MCP-capable assistants such as Codex and Claude to delegate bounded text work — summaries, structured extraction, classification, rewrites, code drafts and log digestion — to locally hosted Ollama models, with conservative routing that keeps small or sensitive tasks with the caller and private-result artifacts for sensitive output.

Enables a single owner to run a self-hosted authenticated MCP server that adds Slack reading and replying, Outlook inbox and calendar lookup, signed Slack event capture, restricted KakaoTalk Windows-agent tasks, appointment tracking, and morning briefings to an existing MCP setup.

Enables an agent to verify its own work mid-task by deliberately breaking each changed line, re-running the test suite in isolated workers, and reporting which breaks no test catches, along with a plan tool that previews which changed lines would be checked.

Enables MCP clients to coordinate local AI model runs by monitoring RAM, system commit, and VRAM, queueing tasks across clients, performing admission checks before inference, renewing leases, completing and releasing runs, and reporting status, history, and health while preserving original generation quality parameters. It also supports controlled start/stop and optional idle application memory reclamation.

Enables read-only monitoring and diagnostics of Apache Flink clusters through a JobManager REST endpoint, including job, checkpoint, backpressure, TaskManager, and multi-cluster overviews with LLM-friendly summaries.

A task-aware transparent proxy that wraps stdio MCP servers (such as git, filesystem, and terminal) and compresses large tool outputs field-by-field using declarative YAML profiles, passing narrow intent queries through untouched. Discarded data is stored in SQLite and can be losslessly retrieved via an injected tool, either in full, by line range, or by grep, while errors are always preserved and failures fall back to the raw result.

Enables building zero-dependency Model Context Protocol servers over the Streamable HTTP transport, exposing typed tools, resources and URI templates, prompts, background tasks, and OAuth-protected endpoints. It can be mounted into FastAPI, Starlette, Flask, Django or any ASGI/WSGI app, or run standalone on the Python standard library alone.

Exposes device discovery, capabilities, observation, logs, task execution and cancellation, and evidence retrieval as tools so MCP clients can run typed mobile automation tasks and inspect proof of completion.

Turns a Frappe site into a remote MCP server that AI clients such as Claude can sign in to with OAuth and use curated, task-level tools registered by installed apps. Every call runs as the signed-in user with that user's permissions, with schema-checked arguments and rollback on failure.

Enables querying a Chatwoot support inbox in plain language—finding conversations, contacts, and conversations needing attention—and, after explicit confirmation, adding private notes or updating assignments, labels, and status without sending customer messages.

Enables plain-language queries about a support desk, ticket and customer investigation, and safe, confirmation-gated ticket assignment and internal notes.

Enables natural-language querying of GLPI IT helpdesk and asset data—tickets, ownership, deadlines, and person-held assets—plus safe follow-up additions and ticket updates that require one-time confirmation before changes are applied.

Enables coding agents to index repositories, retrieve task-relevant code within a token budget, and record facts that are flagged stale when their anchored code changes.

Enables MCP-compatible AI clients to search and view CRM customers, deals, contracts, tasks, calendars, and outbound jobs, and to propose opt-in write actions that require human approval in the CRM.

Enables AI coding agents to automatically compress context at agent-judged task boundaries, write incremental structured session snapshots, and maintain per-fact long-term memory for priority restoration across sessions.