"Zsh" matching MCP servers:

Zero-dependency MCP server that equips text-only LLMs (e.g., DeepSeek) with vision by converting images to text descriptions via OpenAI-compatible vision APIs, exposing an img2text tool with multi-provider fallback.

MCP server for generating and editing images and videos via DashScope/OpenAI-compatible APIs, with tools for text-to-image, image editing, text/video generation, and automatic local download.

Turn a sentence into an image or a short video on your machine.

An MCP server providing vision understanding tools (image QA, OCR, video and document understanding) via OpenAI-compatible endpoints, defaulting to Zhipu GLM vision models.

Enables MCP clients like Codex CLI and Claude Code to delegate tasks to a local DeepSeek Harness (dsh), with native Web transport for visible progress, model selection, and persistent long-running jobs, plus an ACP-compatible transport.

MCP server for recognizing and generating images, videos, and audio via DashScope, with support for large video uploads and local output previews.

Certification lookup for DeepSeek Harness plugins: an agent-facing interface to the dsh-plugin-certification registry, checking a plugin's five reproducible dimensions (manifest compliance, build hygiene, supply-chain Scorecard, release provenance, sandboxed install smoke) before recommending or installing it.

Enables batch translation of arXiv, DOI, or paper-page LaTeX sources into Simplified Chinese and compilation of Chinese PDFs via XeLaTeX, using DeepSeek's deepseek-flash model with glossary, structure validation, resume, and QA.

An MCP bridge that exposes DeepSeek Harness (DSH) web UI's cordis RPC API as stdio tools, enabling any MCP client to manage workspaces, create/resume sessions, send messages, and fetch session stats.

Local-first toolkit with 16 tools for Chinese web-novel (网文) authors: style baselines, foreshadowing and plot-thread tracking, continuity & OOC audits, keyword and sentence-pattern analysis, and offline semantic search. No API key, no network — everything runs on your machine.

Made-to-order data for AI agents via x402 micropayments on Base. Describe a need in plain language, get a custom quote, pay per call. No signup, no API keys. HTTP + MCP transports. 5 tools.

A cognitive scripting language for structured reasoning with LLMs. 12 built-in operations (survey, ground, assert, doubt, contrast, synthesize, etc.) for guiding AI through dialectical analysis, impact assessment, and other thinking patterns.

MCP server that defers non-urgent tasks to off-peak hours, saving ~50% on DeepSeek API costs without blocking users.

Enables voice control of DeepSeek Harness Web via the Xiaozhi assistant, exposing DSH workspaces, sessions, models, files, and settings as MCP tools over WebSocket.

Enables Claude Code to act as a director of persistent DeepSeek Harness (DSH) workers through six tools — dsh_spawn, dsh_wait, dsh_list, dsh_kill, dsh_send, and dsh_answer — using blocking polls instead of push events.

Lets Google Antigravity's agent request tools that DeepSeek Harness validates, authorizes, and executes, covering files, commands, and installed extension tools, with the real result returned over the same ACP prompt.

Enables MCP clients to run materials computation workflows through tools for loading structures, relaxing potentials, screening doped variants, analyzing phonons and elasticity, sampling candidates, and replaying trajectories. It exposes these capabilities over stdio or streamable HTTP, adapting available engines and structure sources to the environment.

MCP server for detecting AI writing patterns in Chinese academic text and providing rewrite guidance to reduce AIGC detection rates.

MCP server bridging ChatGPT to DeepSeek Harness, exposing 13 tools for task submission, status tracking, result retrieval, project/session management, and human-in-the-loop approvals.

Exposes a locally installed DeepSeek Harness as an MCP server so Codex can dispatch real tasks (with the harness's own skills, memory, and workspace) and check its own wiring, returning answers plus resumable session IDs. Every call is recorded in a per-call, auditable JSONL log.