MCP servers that ship as code — a repository or a package you install and run locally (usually over stdio). They have no hosted endpoint, so we cannot probe or grade them: instead each one shows its repository, install command, license, stars and when it was last updated.
Collected from GitHub repositories and npm, PyPI, Cargo, Go, Maven, NuGet and RubyGems packages tagged as MCP servers, the official MCP Registry, Docker's MCP catalog, ModelScope, Baidu MCP World, Cline and other public directories; one entry per repository. A server that also offers a hosted endpoint is listed under Remote instead. These entries are not counted in our reachability, verification or readiness statistics.
-
offload-mcpupdated this year
MCP server that offloads routine AI coding tasks to free LLM APIs
npm MIT -
offload-mcpgithub.com/peterhadorn/offload-mcpupdated this year
MCP server that offloads routine AI coding tasks to free LLM APIs
MIT ★ 6 -
Delegate summarize/classify/extract/triage to a FREE local Gemma-4 cascade via llama.cpp. Go CLI + MCP server; never calls a cloud model. Includes a turnkey setup skill for Claude Code.
go Apache-2.0 ★ 4 -
MCP server that lets Claude Code (or any MCP client) delegate sub-tasks to a local Ollama model
MIT ★ 1 -
Let your main coding agent delegate heavy reading to a second agent, and read back only the answer.
npm MIT ★ 23 -
Claude の軽量な LLM 作業(要約・分類・抽出・簡単な質問)を、ローカルモデルや任意の OpenAI 互換プロバイダ(LM Studio・Ollama・OpenRouter・Grok・OpenAI・Groq・Together…)へオフロードする MCP サーバー | MCP server that offloads light LLM work (summarize, classify, extract, quick asks) from Claude to a local model or any OpenAI-compatible provider — LM Studio, Ollama, OpenRouter, Grok, OpenAI, Groq, Together…
MIT ★ 1 -
Call another MCP client from your MCP client. Offload context windows, delegate tasks, split between models
MIT ★ 28 -
claude-token-saver-mcpgithub.com/BlackFoil/claude-token-saver-mcpupdated this year
MCP server that offloads coding tasks to local LLM (Ollama) to save Claude API tokens
npm Apache-2.0 ★ 3 -
grok-mcpgithub.com/zekaizer/grok-mcpupdated this year
MCP server: offload research, X search, and analysis to SuperGrok via the xAI Responses API
MIT ★ 1 -
code-aidegithub.com/logicia32/code-aideupdated this year
Offload code reading / review / edits to a cheap LLM via MCP — returns concise results to keep your main agent's context lean.
MIT -
An MCP server that offloads cheap work from your cloud LLM agent to a local Ollama model.
pypi MIT -
CodexMCPgithub.com/JohanFJ101/CodexMCPupdated this year
MCP server exposing a query_offload tool so an agent can hand mechanical work to a cheaper model and keep reasoning in context.
-
parallelix-mcpgithub.com/parallelixnetwork/parallelix-mcpupdated this year
MCP server for the ParalleliX Compute API. Offload parallel inference sub-tasks to a distributed GPU network, paid in $PRLX credits.
npm MIT -
Claudemcpgithub.com/snakeying/Claudemcpupdated this year
Codex → Claude 的 MCP 接口,用 Claude Code 做 review / plan / 架构分析。Expose Claude Code as an MCP server so Codex can offload review, planning, and architecture analysis.
MIT ★ 1 -
Claude Code plugin: MCP server that offloads bounded LLM tasks to cheaper local (LM Studio, Ollama, vLLM, llama.cpp) or remote (OpenRouter) models with profile-based configuration and ensemble mode
MIT ★ 1 -
search-meshgithub.com/jamestkelly/search-meshupdated this year
SIMD-accelerated, single-pass codebase intelligence fabric for AI agents. Offload multi-keyword scans and AST context-pruning to native silicon.
MIT ★ 3 -
Step1X-Edit-Dockergithub.com/neosun100/Step1X-Edit-Dockerupdated this year
🎨 Production-ready Docker deployment for Step1X-Edit with intelligent GPU management. Features: UI + API + MCP, lazy loading, instant offloading, multi-language support.
Apache-2.0 ★ 1 -
Serve, size and benchmark open LLMs on one 16 GB GPU: vLLM vs llama.cpp vs Ollama, KV cache and offload planning, MCP server
MIT -
Stdlib-only MCP server that offloads mechanical work from Claude Code to free/cheap LLM APIs (Groq, Gemini, OpenRouter) - ~90% fewer Claude tokens
MIT -
calculus-helper-mcp-demogithub.com/rdwj/calculus-helper-mcp-demoupdated this year
MCP server that offloads symbolic calculus to SymPy — 8 tools (derivatives, integrals, limits, Taylor series, equation & ODE solving, simplification, numeric evaluation) so LLM agents get computed answers instead of hallucinations. Built on FastMCP 3.x.
NOASSERTION -
MCP server for offloading tasks to local LLMs via LM Studio SDK - now with context-aware prompts, diagnostics, and bug fixes
npm MIT -
lethe-engineeringgithub.com/JesusGarcia9009/lethe-engineeringupdated this year
Context garbage collector for long-running LLM agents - offload & recall to save tokens in Claude Code and Codex via MCP.
pypi Unlicense -
local-llm-delegation-mcpgithub.com/mrrodriguez/local-llm-delegation-mcpupdated this year
Intelligent task delegation for AI agents: Offload the "easy" work to your local machine and save premium tokens for the hard stuff.
★ 1 -
A repo investigator for Claude Code and Codex. Offload codebase investigation and get up to 2.7× as much work from the same limits.
npm MIT ★ 37 -
Universal MCP server for LLM context offloading: AST skeletons, BM25 symbol search, surgical patches, smart log pruning and distilled external schemas. ~90% fewer tokens. Windows/macOS/Linux, zero compilation.
-
Universal MCP server for LLM context offloading: AST skeletons, BM25 symbol search, surgical patches, smart log pruning and distilled external schemas. ~90% fewer tokens. Windows/macOS/Linux, zero compilation.
-
llm-cost-router-mcpupdated this year
MCP server that advises LLM model routing & cost BEFORE the call — estimate_cost, recommend_model, should_offload_local. Pure logic, no API keys, $0 to run.
npm MIT -
local-dev-mcpupdated this year
Local-first MCP server that offloads code safety, boilerplate generation, dependency evaluation, project indexing, and task delegation to local LLMs via oMLX or LM Studio.
npm MIT -
Quiet MCP executor for Claude Code and Codex. Offload edits, tests, refactors, and commits to cheaper local or flat-rate models.
npm MIT ★ 4 -
MCP server that ranks log patterns by volume and cost and plans compact, tier down or offload per pattern. Runs locally; no log data reaches log10x.
go MIT ★ 1 -
freeloadergithub.com/1977-code/freeloaderupdated this year
Claude Code MCP server + skill that offloads cheap, bulk subtasks to free local LLMs via a self-hosted freellmapi router. Save your Opus/Sonnet tokens.
MIT -
alt-llm-plannergithub.com/VikAnalytics/alt-llm-plannerupdated this year
MCP tool for Claude Code & Cursor: offload architectural planning Q&A to a side-terminal LLM (Gemini today, others next), return a dense markdown blueprint to Claude. Saves context window + token cost.
MIT -
Zero-dependency 3-tier MCP router for Claude Code & Google Antigravity that offloads file scouting and test log triage to a local GPU with automatic cloud failover.
Apache-2.0 -
cheaplanegithub.com/millennialdreamer/cheaplaneupdated this year
Offload replaceable grunt work from your premium AI agent to cheap models via MCP. Keep your subscription on the main thread; billing is physically isolated. Auto-routing + savings ledger + one-command setup. MIT.
pypi MIT ★ 1 -
Peripheral Evaluation & Execution Proxy: The open-source token shield for AI coding harnesses that offloads mobile visual perception, coordinate grounding, micro-loops, and noisy log analysis to local models.
npm Apache-2.0 -
MCP servers that add verification, code intelligence, context packing, memory, sub-task offload, supply-chain checks and deep research to the OpenAI Codex CLI on any model backend.
-
A Claude Code plugin that offloads the heavy lifting — reading, searching, and batch-editing code — to your local LLM. Claude does the thinking; your local model does the grunt work.
MIT -
Full Page Shot: Microsoft Edge extension + MCP server for AI agents. Capture full-page screenshots (PNG/JPEG) of any page. Saves agents tokens by offloading screenshot capture. TypeScript, esbuild, Manifest V3. https://albertovasquezr.github.io/full-page-shot/
MIT -
Offloads large-context work like file exploration, indexing, summarization, deep research — to Google’s free Antigravity CLI (agy), the terminal coding agent powered by Gemini models. Claude orchestrates and reasons; Antigravity handles the heavy lifting.
MIT ★ 1 -
CodeBraingithub.com/Tschonsen/CodeBrainupdated this year
An MCP server that offloads bulk work from Claude Code to a local LLM (Qwen2.5-Coder 14B via Ollama). Claude stays the reasoner; the local model takes the bulk drafting, polishing, and explanation work to free up rate limits and context budget.
MIT ★ 1 -
Launch open-source chat, voice, image and video models on your Mac (Linux in preview). Rust core, RAM limits with Run anyway and SSD offload, web access, computer-use benchmarks, OpenAI-compatible API and MCP server.
MIT ★ 2 -
hermes-mcpgithub.com/SPhillips1337/hermes-mcpupdated this year
Hermes MCP is a Model Context Protocol (MCP) server that acts as a bridge for the Hermes Agent. It enables AI tools, autonomous agents, and IDEs (like Claude Desktop) to leverage Hermes as a powerful sub-agent for offloading complex tasks, performing deep research, or architectural planning.
MIT ★ 2 -
Claude-led collaborative development framework: Claude Code plans, implements, reviews, and verifies; OpenAI Codex is an optional engine for offloaded coding tasks. Spec-driven, with traffic-light verification + academic report generation. Built on Pimzino/spec-workflow-mcp.
GPL-3.0 ★ 2 -
Suber-Agent-Teamgithub.com/stephenpham68/Suber-Agent-Teamupdated this year
MCP server that offloads Claude Code sub-agents to a cheap model (Haiku / any OpenAI- or Anthropic-compatible API) - run 20-100 parallel agents for fan-out, map-reduce & research without draining your Opus / Claude Max quota.
MIT ★ 3 -
figma-edit-mcpno update data
Connect AI assistants to Figma via Model Context Protocol to read designs, create and modify elements, and manage design systems programmatically. This plugin empowers your AI assistant to become an active design co-pilot, executing updates and modifications Safer, Cleaner, and Faster than manual entry. By offloading repetitive, error-prone manual operations to automated systems, designers are freed to focus entirely on creative decision-making and product strategy. ### Core Principles * **🛡️ Safer**: The server enforces robust programmatic checks and validations that exceed the Figma Desktop app's default protections. For instance, it prevents deleting variables currently in use to avoid dangling references. By validating all writes against a locked editable scope and strict name checks before execution, it protects files from human error and AI hallucinations. * **✨ Cleaner**: System-driven, programmatic operations guarantee that no layers, text nodes, or components are skipped or forgotten during complex updates. Every design file stays consistent and conforms exactly to your design system. * **⚡ Faster**: Bulk edits like propagating instance overrides, batch-updating copy, or restructuring auto-layout frames are handled via high-performance APIs. Hours of tedious, click-heavy manual work are completed in seconds. ### How it Works (Prerequisites) This server runs locally on your machine. To operate, it requires: 1. The Figma Desktop App to be open. 2. The `Figma Edit` plugin to be active on the file you want to modify (listening on local port `3845`).
smithery -
Standalone Model Context Protocol (MCP) Server providing real-time cognitive research registry queries, EU AI Act compliance auditing, and adversarial human-AI interaction benchmarks.
npm MIT