Machine learning MCP servers

A server here trains, fine-tunes, evaluates or hosts machine learning models.

Not here: servers that merely use AI to do something else.

463servers
239 remote224 localTagged by Odel's AI, which can make mistakes
By GitHub stars, at most two per publisher
  1. Number 1: Pollinations@pollinations/pollinationsAccess Pollinations models and API capabilities through agent tools.Not assessed yet.A remote server; it needs your API key.5,183 GitHub stars.
  2. Number 2: Opik@comet-ml/opik-mcpRead and score Opik traces, experiments, datasets and prompts from your AI assistant.3% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; you sign in with your account.220 GitHub stars.
  3. Number 3: Stats Compass@oogunbiyi21/stats-compass50+ pandas-powered tools for data loading, cleaning, visualization, and ML workflows7% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; you sign in with your account.17 GitHub stars.

Best fit

247 made for this, ranked by starsRemoteLocal
  1. FitLLM @fitllm/fitllmWill this LLM fit on your GPU, multi-GPU rig or Mac? Exact VRAM & KV-cache math. Read-only.3% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; no sign-in.8 GitHub stars.
  2. Nonobench @mauricekleine/nonobenchAn open-source benchmark of how well LLMs solve nonogram puzzles, from 5x5 to 20x20.Not assessed yet.A remote server; no sign-in.7 GitHub stars.
  3. Disco @leap-labs/discovery-engineFind novel, statistically validated patterns in tabular data — hypothesis-free.5% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; no sign-in.7 GitHub stars.
  4. Hologram Model Hub @uor/model-hubFind open models and get download links with the SHA-256 each file must have.5% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; no sign-in.5 GitHub stars.
  5. Createya — AI Image & Video Generation @createya-ai/createya-mcp100+ AI models: FLUX, Sora, Veo, Kling, Runway, Suno. OAuth or Bearer. No VPN. RU billing.15% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; you sign in with your account.4 GitHub stars.
  6. CompletionKit @completionkit/evalsPrompt evals over MCP: run a prompt on your dataset, score each output 1-5 with an LLM judge.4% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; no sign-in.3 GitHub stars.
  7. Palimpsest — censorship, China economy and model-eval observatory @beepboop2025/palimpsestRights-aware censorship, China-economic status, and tamper-evident AI evaluation tools.5% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; no sign-in.3 GitHub stars.
  8. Tenzro Ledger MCP @tenzro-network/tenzroTenzro Ledger MCP: wallet, identity, payments, inference, staking, bridges, verification, agents.10% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; no sign-in.3 GitHub stars.
  9. Nimbus BCI @nimbusbci/nimbus-mcpAI agents build, train, and analyze BCI/EEG pipelines: data, models, experiments, live sessions.Not assessed yet.A remote server; no sign-in.2 GitHub stars.
  10. NLP Tools - Sentiment, NER, Toxicity & Language Detection @fasuizu-br/nlp-toolsToxicity, sentiment, NER, PII detection, and language identification tools5% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; no sign-in.2 GitHub stars.
  11. DEEPBOM @junhwan-kwon/deepbomBrowser-local and CLI static evidence for deployed AI model artifacts.4% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; no sign-in.2 GitHub stars.
  12. WhitePact @guruprasath-annadurai/whitepactAI governance MCP server: trust scoring, guardrails, bias/hallucination detection, compliance.10% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; you sign in with your account.2 GitHub stars.
  13. Image Tools - Background Removal, Upscaling & Face Restoration @fasuizu-br/image-toolsBackground removal, 4x upscaling, and face restoration via GPU6% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; no sign-in.2 GitHub stars.
  14. VerifiMind PEAS - RefleXion Trinity @creator35lwb-web/verifimind-genesisX-Z-CS AI validation with auditable reasoning. 8 active tools; 5 in maintenance. v0.5.6414% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; no sign-in.2 GitHub stars.
  15. tokcalc MCP Server @stevecrates489-commits/tokcalcLLM serving capacity planner: VRAM, KV cache, GPU topology, latency, and cost.6% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; no sign-in.1 GitHub stars.
  16. Zhijiangyun Cloud (智匠云) @artibot/zhijiangyunRobot embodied-AI cloud for GPU training, inference, benchmarks, and robot data collection.7% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; it needs your API key.1 GitHub stars.
  17. BridgeNode MCP @bridgenode-ai/bridgenode-mcpBridgeNode — x402 pay-per-request AI inference. OpenAI-compatible API + MCP, Solana USDC, gas-free.12% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; no sign-in.1 GitHub stars.
  18. Xplainable @xplainable/xplainable-mcp-serverTrain, explain, optimise and deploy transparent glass-box ML models via workflow tools.6% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; you sign in with your account.1 GitHub stars.
  19. Gpusmarket @jyswee/gpusmarketRent real GPUs from the terminal. Per-second billing, $0.50 minimum. stdio MCP server.8% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; no sign-in.1 GitHub stars.
  20. RunComfy MCP @runcomfy-com/runcomfy-mcpRemote MCP for RunComfy: ComfyUI deployments, hosted models, LoRA training. 31 tools.7% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; you sign in with your account.1 GitHub stars.
  21. Ephemeris Time-Series Forecasting @cascade/ephemerisProbabilistic time-series forecasts from zero-shot foundation models: routed, single or ensembled.5% chance this is malicious. Assessed by AI, which can make mistakes.A remote server; it needs your API key.

Also touches on Machine learning

106 where it is one part of a bigger job