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.
-
Multi-agent pharmaceutical supply chain optimizer using OECD Pharma data (19.6M rows), World Bank financing, and reinforcement learning with MCP tool orchestration
-
temperagithub.com/anvanster/temperaupdated this year
MemRL: MCP Server for Self-Evolving Agents via Reinforcement Learning on Episodic Memory
npm Apache-2.0 ★ 6 -
opencortex-memorygithub.com/stardustvision/opencortex-memoryupdated this year
OpenCortex MCP server and CLI for AI agent memory with reinforcement learning and multi-tenant isolation.
npm MIT -
QuantiaBayes Financial MCP. A Research-Grade Financial MCP implementing Hierarchical Bayesian Inference (HMC) and Reinforcement Learning (PPO) for risk assessment and autonomous decision making.
-
MCP server that personalizes LLM replies for ChatGPT, Claude and Grok. A reinforcement learning bandit learns each user's preferred length and format in real time, with A/B testing to measure impact. Live at claire.pm.
-
Build an autonomous agentic AI system that detects fraud in banking transactions using reinforcement learning for hyperparameter optimization, integrated with MCP servers for tool access and data management.
★ 1 -
Self-improving skill memory for coding agents: learn, recall, reinforce, decay — with provenance on every memory and trust only you grant. Local SQLite, no API key, no cloud.
pypi Apache-2.0 ★ 11 -
GridLab simulates EV charging on a synthetic Novi Sad power grid using OpenStreetMap data and pandapower. Compare charging optimization, grid capacity, reinforcement learning, and Codex/MCP workflows. Schneider Electric EV Days hackathon project.
MIT -
Self-hosted Compound AI System for sovereign environments. Features token-saving complexity routing, deterministic MCP tools for math/logic, configurable expert templates, causal/reinforcement learning, agentic loops, and a federated Neo4j GraphRAG.
Apache-2.0 ★ 12 -
RL & control-aware MCP memory system for structured experiment tracking, theory audit, and safe pattern reuse.
MIT -
gym-mcp-servergithub.com/agentring/gym-mcp-serverupdated this year
Expose any Gymnasium environment as an MCP server
pypi MIT -
An agent memory server that learns from feedback which memories are worth returning. Go, with Python/LangChain/MCP adapters.
MIT -
Multi-agent RL pursuit engine over FastMCP: Dec-POMDP zero-sum game, Q-learning hybrids, natural-language state negotiation, Gmail API telemetry.
-
Probe RL environment verifiers for false passes and false fails
pypi Apache-2.0 ★ 1 -
A learned latent world model an agent plans with, over MCP — JEPA-style next-latent prediction, latent-space MPC, no pixel reconstruction.
MIT -
experiment-auditgithub.com/sreedharshan-gj/experiment-audit-mcpupdated this year
MCP server for auditing ML experiments (W&B): confounded ablations, training-curve pathologies, and misleading sweep conclusions.
pypi MIT -
Self-learning issue memory for Codex using MCP. Automatically captures reusable bug fixes and reuses them across sessions with local-first, low-token SQLite storage.
MIT -
Parlaygithub.com/sh4shv4t/Parlayupdated this year
Parlay — RL negotiation environment where LLM agents learn game theory, theory-of-mind, and strategic bluffing through self-play. OpenEnv compliant · MCP ready · GRPO-trained on Qwen2.5.
★ 1 -
MCP server for stochastic algorithms and probabilistic decision making
npm MIT -
openclaw-hackathongithub.com/Trolleroof/openclaw-hackathonupdated this year
Apollo Labs RL run orchestrator: FastAPI PPO training, eval baselines, rollout GIFs, canonical RunReports, AgentMail delivery, Nia lessons, Next.js dashboard, and MCP tools.
★ 2 -
long-term-memorygithub.com/ihwooMil/long-term-memoryupdated this year
Intelligent memory MCP server for AI assistants — auto-save, semantic search, RL-powered retrieval, and knowledge graphs
pypi MIT -
tuneforgegithub.com/JustVugg/tuneforgeupdated this year
An MCP server for dataset generation, fine-tuning, RL, and evaluation of LLMs — directly from your coding agent.
MIT ★ 17 -
Python SDK and MCP server for Drums: create, act on, fork, reset and verify isolated desktops for computer-use agents
Apache-2.0 -
jar-elgithub.com/ProfEngel/jar-elupdated this year
Personal Semantic OS (S-OS) & Digital Twin. Features "Self-Baking" memory, hybrid architecture, and agentic workflows via the Model Context Protocol (MCP).
★ 4 -
a knowledge-graph and RAG system that extracts requirements into neo4j, queryable through an ollama and MCP chat client with RL tuned retrieval strategies.
-
MCP Server implementing Fractal-RL–THESIA framework with Animal Controller + Large Information Models
npm MIT -
Gymnasium-style RL framework for LLM agent training — MDP environments, three-layer process reward & SFT/DPO/GRPO policy optimization. CLI + MCP ready.
★ 4 -
Hexmend: an agent gym where humans decide what matters and agents prove the smallest repair, through seven WebMCP tools on one executable spell graph.
MIT -
Governed, self-improving multi-agent swarm for 5G network operations on AWS: Bedrock agents, MCP tools behind AgentCore Gateway, Cedar policies, human approval, RL + federated learning on a digital twin. Finalist, AWS x FYUZ Breaking Barriers 2026 (team project).
-
@mseep/thinking-patternsgithub.com/emmahyde/thinking-patternsupdated this year
MCP server combining systematic thinking, mental models, debugging approaches, and stochastic algorithms for comprehensive cognitive pattern support
npm MIT -
MCP server combining systematic thinking, mental models, debugging approaches, and stochastic algorithms for comprehensive cognitive pattern support
npm MIT -
CIEL Aletheia: Sub-microsecond (1.09µs) in-memory fuel-metered verification engine for AI reasoning, MCP tool-calls, and RL rollouts (Apache 2.0)
★ 1 -
mcp-app-servergithub.com/sharathStack/mcp-app-serverupdated this year
FastMCP server integrating a stateful app into RL sandbox environments — populate/snapshot/restore hooks with content-addressed state digests, Docker packaging, and a 51-test suite spanning the app, state, and protocol layers.
-
An open-source robotics workshop for agents: simulate complete robots on high-fidelity physics, build digital twins, make the parts on CNC, laser and 3D-printer twins, connect simulation to real robots, and let AI agents program, test and debug it all, just by talking to it.
Apache-2.0 ★ 197 -
Runnable sample of the OpsDesk API edition: care-home rostering under UK or German labour law for tool-use agents (HTTP/MCP), scored against a solver reference. Claude Opus 5.5 solved 0 of 13 hard tasks.
NOASSERTION -
MCP server that lets AI agents drive a live, persistent Isaac Sim session: load any robot, set joint targets and poses, tune gains live, run joint range tests, take screenshots, and launch Isaac Lab training runs. Kit stays up between calls, so each experiment costs a frame instead of a full relaunch.
pypi MIT