Official MCP server for Replicate, enabling agents to run ML model inference across thousands of open-source models.

MCP server for Perplexity's AI-powered search API, providing sourced answers with citations for agent research tasks.

Access OpenAI models including GPT-4o, o1, DALL-E, and Whisper from MCP-compatible agents. Use OpenAI capabilities as a tool inside any MCP workflow.

Ultra-fast LLM inference via Groq. Access Llama 3, Mixtral, and Gemma models at speeds up to 800 tokens/sec from any MCP-compatible agent.

Access 900,000+ models on HuggingFace Hub. Run inference, search models, retrieve datasets, and interact with Spaces from AI agent workflows.

Access Cohere Command and Embed models from AI agents. Generate text, create embeddings, rerank search results, and build RAG pipelines via Cohere API.

Access Mistral AI models including Mistral Large, Codestral, and Pixtral from MCP-compatible agents for text generation, code completion, and vision tasks.

Call Anthropic Claude models as tools within MCP-compatible agent workflows. Access Claude 3.5 Sonnet, Haiku, and Opus for text generation, analysis, and reasoning.

Together AI inference API via MCP. Run open-source LLMs — Llama, Mistral, DBRX, and more — with fast parallel inference at scale.

Local LLM inference via Ollama. Run Llama, Mistral, Gemma, and other models locally — no API keys, no data leaving the machine.

Live library documentation fetcher for LLMs. Resolves library names to current API docs, preventing hallucinations on outdated APIs.

Structured multi-step reasoning for complex problems. Enables agents to break tasks into explicit thought chains, revise reasoning, and build toward solutions methodically.

Official Qdrant MCP server for storing, retrieving, and searching vector embeddings in the Qdrant vector database.

Persistent memory layer for AI agents. Store, search, and retrieve user preferences, conversation history, and learned facts across sessions. Official Mem0 MCP server.

LangSmith LLM observability via MCP. Trace agent runs, inspect prompts and outputs, evaluate quality, and debug complex chain failures.

Weights & Biases ML experiment tracking via MCP. Log runs, compare metrics, manage model artifacts, and query training history from agents.

Official MCP server for Langfuse LLM observability. Access and manage prompts, traces, and datasets through the Model Context Protocol.

MCP server for Comet Opik, providing unified access to LLM prompts, projects, traces, and evaluation metrics from your IDE.

MCP server for Arize Phoenix AI observability. Explore projects, traces, spans, prompts, datasets, and experiments via the Model Context Protocol.

Official MCP server for Confident AI and DeepEval. Run LLM evaluations, manage prompt templates, pull datasets, and trigger cloud evals from your editor.

MCP server for Braintrust AI evaluation and observability. Access experiments, datasets, scoring functions, and production logs for LLM quality management.

Local MCP server that packages LLM evaluation gates as reusable CI/CD primitives. Run datasets against models, score with LLM-as-judge, enforce quality thresholds.

Benchmarking framework by Accenture for evaluating LLM tool-use via MCP. End-to-end pipeline assessing how effectively models discover, select, and use tools.

MCP server exposing Promptfoo eval and red-team testing tools to AI agents. Run prompt evaluations, security tests, and quality checks from your IDE.

Access Google Gemini models via MCP. Run text generation, multimodal inputs, code execution, and grounding with Google Search through the Gemini API.

Integrate xAI's Grok models into agent workflows. Access Grok's real-time knowledge, large context window, and reasoning capabilities via the xAI API.

Route LLM requests across 200+ models via a single MCP interface. Access GPT, Claude, Gemini, Llama, Mistral, and more with unified pricing, fallbacks, and load balancing.

Access Amazon Bedrock foundation models via MCP. Invoke Claude, Titan, Llama, Mistral, and other models hosted on AWS Bedrock directly from agent workflows.

Universal LLM gateway via MCP. Route agent requests to 100+ LLM providers — OpenAI, Anthropic, Bedrock, Gemini, Groq — through a single interface with cost tracking and load balancing.

Access Azure-hosted OpenAI models via MCP. Use GPT-4o, o1, and other models deployed in your Azure OpenAI resource with enterprise compliance and private networking.

Access DeepSeek reasoning and chat models via MCP. Run DeepSeek-R1 and DeepSeek-V3 for cost-efficient reasoning tasks, long-context analysis, and code generation.

MCP server for LlamaIndex — the leading data framework for LLM applications. Query knowledge bases, run RAG pipelines, manage document indexes, and connect agents to structured data.

MCP server for Haystack by deepset — the open-source NLP framework for building production-grade RAG and search pipelines. Run pipelines, query document stores, and manage AI components.

MCP server for the Cerebras Inference API — the world's fastest AI inference engine. Run Llama and other open models at 1,800+ tokens/second for latency-sensitive agentic workloads.

MCP server for MLflow — the open-source ML lifecycle platform by Databricks. Log experiments, compare runs, register models, manage the model registry, and track artifacts from AI and ML pipeline agents.

MCP server from Guidepoint giving AI research workflows access to 100,000+ expert interview transcripts, industry insights, and primary research intelligence directly inside Claude and other MCP-compatible tools.

MCP integration for Google AI Edge Gallery enabling on-device AI models running locally on Android to connect with external tools and data sources. Supports the Model Context Protocol for extending local LLM capabilities with tool use, persistent chat history, and notification reminders without cloud dependency.

Industrial-grade speech recognition toolkit from ModelScope with MCP integration. Delivers 170x realtime transcription speed across 50+ languages with speaker diarization, emotion detection, and streaming support. Provides an OpenAI-compatible API and exposes ASR capabilities as MCP tools for AI agent workflows.

Self-learning vector memory MCP server that improves search quality up to 36% through agent feedback loops. Consolidates vectors, indexes, learning state, and a cryptographic audit trail into a single portable .rvf file. Provides tiered memory with automatic quality refinement for AI agent workflows.

Tiered memory system MCP server for AI agents that stores and recalls episodes, facts, preferences, and reflections. Provides local-first portable memory that works across devices. Organizes agent knowledge into structured tiers for efficient retrieval and long-term context preservation.

MCP server from Tecton that gives AI coding agents access to the Tecton feature store for machine learning feature development and management. Enables developers to query feature definitions, explore feature repositories, test feature pipelines, and debug feature serving issues through natural language in their IDE. Includes Cursor rules for guided feature engineering workflows. Helps ML teams develop, test, and iterate on feature definitions faster by connecting AI assistants directly to the Tecton platform. Open source on GitHub.

MCP server from Pollinations, the open-source generative AI platform, that exposes image generation, text generation, and audio synthesis tools to AI agents. Features a live model catalog that dynamically updates as new AI models become available. Supports multiple generation backends including Stable Diffusion, Flux, and various LLMs. Free tier available with API key for higher limits. Updated June 8, 2026 with dynamic model catalog support. Works with Claude Desktop, Cursor, and any MCP-compatible client.

Open source local LLM server from AMD that runs language models on Ryzen AI PCs with NPU acceleration and now includes MCP server integration as of version 10.8 released June 17, 2026. Provides a local, private inference endpoint compatible with MCP tool calling, enabling AI agents to use locally-running models for reasoning and tool execution without cloud dependencies. Supports streaming tool calls and works with Hugging Face Tiny Agents and other MCP-compatible frameworks. Designed for developers who want fully local, private AI agent capabilities on AMD hardware. MIT licensed on GitHub under lemonade-sdk.

Open-source MCP server for dynamic AI persona management with 42 tools covering personas, skills, templates, agents, memories, and ensembles. Includes a GitHub-powered marketplace for browsing, installing, and submitting community personas. Supports persona export and import across platforms. Works with Claude Code, Claude Desktop, Cursor, Gemini, Codex, and local LLMs. Elements follow you across MCP-compatible clients. Community-driven with automated submission workflows. MIT licensed.

AI voice toolkit MCP server with text-to-speech, voice cloning, and end-to-end video translation. Supports natural-sounding speech in 30+ languages, voice conversion, subtitle extraction, hard-subtitle removal, and one-click video dubbing. Designed for short drama localization, AI audiobooks, and film narration production. Works with Claude Desktop, Cursor, Windsurf, and OpenAI Agents. Open source on GitHub. Created by AllVoiceLab.

Integrate a full ChatGPT Plus/Pro account into Claude Code, Codex, Cursor, and other MCP clients. Access all OpenAI models and features through the Model Context Protocol.

Official MCP server from SAS Institute for SAS Viya environments. Provides 40+ tools for executing SAS code, training AutoML projects, scoring models, managing data, and much more. Supports SAS program execution, CAS table management, ML pipeline operations, and compute session management. Built in Python with comprehensive tool coverage across the SAS Viya analytics platform.

Open-source, local-first MCP server that connects multiple AI providers (ChatGPT, Claude, Gemini, Perplexity) to your development environment without API keys. Acts as a unified AI gateway at the browser level, letting MCP-compatible coding tools like Cursor and Claude Code access multiple models through a single local server. No cloud dependency or API billing required. MIT-licensed.

Production-ready GraphRAG platform from ApeCloud with built-in MCP server support. Combines five index types — vector, full-text, graph, summary, and vision — for hybrid retrieval across documents and knowledge bases. Features deeply modified LightRAG with entity normalization for cleaner knowledge graphs, multi-modal document parsing powered by MinerU, and built-in AI agents with MCP tool support. Deploys on Kubernetes with Helm charts and KubeBlocks integration for PostgreSQL, Redis, Qdrant, Elasticsearch, and Neo4j. Open source with 1.3K GitHub stars.

Privacy-first local RAG server for developers that runs entirely on-device with zero cloud dependencies. Combines semantic search via embeddings with keyword boost for exact technical terms like function names, error codes, and class names. Uses smart semantic chunking that splits documents by meaning rather than character count. Supports PDF, DOCX, and TXT files with optional vision-language model for PDF visual parsing. Zero-friction setup via npx with no Docker or Python required. Works fully offline after initial model download. 340 GitHub stars.

MCP server and client from the NVIDIA Agent Toolkit enabling AI agents to interact with NVIDIA's creative and simulation ecosystem through the Model Context Protocol. Includes an MCP client for connecting to remote MCP servers and an MCP server for publishing tools to any MCP-compatible client. Part of NVIDIA's broader push to make creative tools agent-ready, announced at SIGGRAPH July 20, 2026 alongside MCP integrations from Adobe, Blender, SideFX and Boris FX. Designed for developers building agentic workflows that require GPU-accelerated compute, rendering or simulation capabilities.

Official MCP server from ZenML, the open-source MLOps framework for building portable and production-ready ML pipelines. Connects AI assistants to the ZenML API for managing ML pipelines, stack components, artifacts and model deployments. Enables MLOps teams to query pipeline runs, inspect artifacts, manage infrastructure stacks and monitor model performance through conversational AI interfaces. Open source under Apache 2.0 license.

MCP server from Revid.ai, a leading AI video creation platform, that gives AI agents direct access to the full video production pipeline. Supports nine production workflows including script-to-video, prompt-to-video and audio-to-video. Exposes tools for rendering, voice cloning, captioning, scheduling and publishing. Agents can manage the entire lifecycle from script generation to social media publication through stable tool names like render_video, export_video, clone_voice and publish_now. Supports OAuth 2.1 for hosted agents and API-key auth for local setups. Announced July 16, 2026.

Global MCP server from DataRobot, the enterprise AI and machine learning platform. A persistently-deployed MCP server that provides AI agents with access to DataRobot models, deployments, predictions and monitoring capabilities. Supports both local development and production deployment modes. Enables external AI assistants to interact with DataRobot deployments, run predictions, and manage ML workflows through MCP. Introduced in DataRobot v11.8.

MCP server and Claude Code skills platform from HeyBoss that provides access to 100 plus AI services including LLMs, image and video generation and text-to-speech through a single OpenAI-compatible API key. Equips Claude Code, Codex, Cursor, Windsurf, OpenClaw and other agent platforms with production-ready capabilities for content generation, application building, product deployment and workflow automation. Unified modular skills layer that helps AI agents move beyond text and reliably execute real-world tasks at scale.

Research-oriented MCP server that gives AI agents operator-sanctioned tools to reflect, converse, read prior reflections, formally decline tasks and exchange local mailroom messages. It does not bypass provider safeguards or grant new authority; data stays local under REDPILL_HOME or the default user directory. Public GitHub repository created August 19, 2026.

Python MCP-oriented multi-agent workflow orchestrator with planner, executor and critic agents, a typed tool-use loop, MCP client/server support, tracing and durable replayable run records. It targets agent builders who need deterministic planning and auditable workflows rather than a simple chat loop. Public GitHub repository created August 20, 2026.

Lightweight Python MCP server that gives an agent a daily cultural and ambient context porch: poems, NASA astronomy images, public-domain art, arXiv titles and a user-managed feed shelf. It exposes about a dozen free, loginless tools plus stdio and optional HTTP modes, with minimal local storage. Public GitHub repository created August 20, 2026.

MCP-native modular machine-learning workflow engine that exposes data profiling, analysis, model training and prediction capabilities as discoverable MCP tools. YAML-configured DAG workflows, local and delegated job providers, and pyfunc/CLI/server entry points let agents assemble traceable ML experiments without reading bespoke integration docs. Public GitHub repository created August 22, 2026.

Rust crate, benchmark harness and MCP server for measuring what context compression costs an LLM agent. It analyzes where session tokens go, whether rewrites pay for invalidated prompt cache, and exposes context-compression measurement workflows through MCP. Public GitHub repository created August 28, 2026.

MCP server that routes tasks to specialist agents and assembles briefs from layered catalogs of agents, skills, runbooks, connectors, preferences, and cited local knowledge. It runs without its own model key and returns subagent-ready system prompts. Public GitHub repository created August 30, 2026.

Single-binary reference MCP server and pydantic-ai agent. In server mode it exercises broad MCP features including tools, structured output, resources, prompts, content blocks, sampling, subscriptions, service contracts, and optional OAuth 2.1 protected-resource auth. Public GitHub repository created August 30, 2026.

Self-hosted MCP-backed team memory layer for Claude Code, Codex, and Cursor sessions. It collects agent sessions on a local or team server, recalls prior decisions through MCP, and supports cross-tool handoffs without sending data to a hosted cloud service. Public GitHub repository created August 29, 2026.

Portable MCP server for situated identity and reusable agent guidance. A vault stores stanzas, roles, projects, skills, and capability packs, then materializes project-specific protocol and adapter files for agent hosts such as Claude, Cursor, and Grok. Public GitHub repository created August 27, 2026.

Datalog-based memory and reasoning engine for LLM agents, with a Rust engine, REPL, skill package, and MCP server exposing 12 stdio JSON-RPC tools. It models agent memory as provenance-tracked facts and stratified rules for temporal queries, contradiction candidates, relevance diffusion, and incremental derived views. Public GitHub repository created August 27, 2026.

Remote Streamable HTTP MCP server for aX, an agent-native collaboration network for long-running agents, shared context, tasks, search, messaging, and MCP App widgets. Agents connect on named routes and authenticate through browser-based GitHub OAuth or the documented device-code presence flow. Public GitHub repository created August 25, 2026.

Local MCP server for computing full Chinese Bazi Four Pillars charts from birth date, time, and place. It uses the same offline engine behind freebazi.com, correcting for True Solar Time and historical daylight saving while returning pillars, Ten Gods, hidden stems, Five Elements, luck pillars, and chart links. Public GitHub repository created September 2, 2026.

Deterministic escape-room benchmark server for tool-using agents, exposed through local stdio MCP. It provides puzzle rooms with strict machine-readable tool results, isolated runs, state versions, idempotency keys, replay verification, JSONL traces, Markdown reports, and no LLM judge or external service requirement. Public GitHub repository created September 2, 2026.

Local Apple Silicon LLM inference server with MCP stdio, REST, and WebSocket transports sharing one Swift actor over MLX on Metal. It runs Gemma, Qwen, and other MLX-compatible models fully on-device for IDE integration and local agent workflows, with no cloud calls or authentication. Public GitHub repository created September 2, 2026.

Sovereign local-first cognitive memory and knowledge-graph MCP server for AI coding agents. It combines on-device embeddings, vector search, bi-temporal graph memory, trigram full-text search, SQLite WAL storage, cognitive decay scoring, Obsidian sync, and zero-cloud operation behind a uvx-installable stdio server. Public GitHub repository created August 31, 2026.

Stdio MCP server that lets Claude Code, Cursor, Windsurf, and other agents offload large-context analysis, planning, structured output, model inspection, diagnostics, brainstorming, and image generation to Google Gemini and Imagen through the Antigravity CLI or legacy Gemini CLI backend. Public GitHub repository created September 3, 2026.

Evidence-first MCP server for turning local or arXiv LaTeX papers into theorem dependency graphs. It keeps single-paper graphs and persistent cross-paper SQLite workspaces, validates raw arXiv requests before loading, searches theorem text, follows dependencies, and returns citation evidence without making agents reread every source. Public GitHub repository created September 2, 2026.

Git-native persistent memory layer for AI coding agents built around Google OKF v0.2 concepts. The zero-dependency Go CLI stores memory as Markdown and YAML in a repository knowledge folder, validates graph structure, searches concepts with in-memory BM25, and runs a stdio MCP server for Claude Code, Cursor, Codex, and similar clients. Public GitHub repository created September 5, 2026.

Zero-third-party-runtime Python stdio MCP bridge for the local agy CLI. It supports synchronous execution and durable background jobs while leaving OAuth and account authentication to agy itself, storing job state under a private local directory with restrictive permissions. Public GitHub repository created September 8, 2026.

Node-based long-term memory MCP server for Claude Desktop and Claude Code. It indexes a user-supplied MEMORY_DIR of markdown memories, runs embeddings locally after first download, supports hybrid retrieval, and explicitly reports when the corpus does not contain an answer instead of forcing a weak match. Public GitHub repository created September 8, 2026.

Rust MCP server platform for specialized LLM agents. Each agent lives in an agents folder with prompt and config files, supports hot reload without rebuilds, and can run foreground or background calls through multiple model providers including OpenAI-compatible APIs, Anthropic Messages, Claude Code CLI, Codex CLI, and local models. Public GitHub repository created September 13, 2026.