App Catalog

AI Stack

AI Model Providers for AI Agents

OpenAI. Anthropic. Google Gemini. Mistral. Cohere. Unified programmatic access to frontier LLMs. We normalize inference APIs, handle token lifecycle, and manage rate-limiting, so your agents can route payloads without hardcoded vendor logic.

Curated by the Vinkius team -- 5 MCP servers reviewed, tested, and ready to connect. Create a free account and start in seconds -- no infrastructure or code needed.

Ask AI about these MCP Servers

Open in ChatGPT Open in Claude Open in Perplexity

OpenAI MCP Server

01 MCP Server

OpenAI MCP Server

The default starting point. Access GPT-4o, o3-mini, and the embeddings that power half the internet.

OpenAI defines the standard inference baseline for complex agentic workflows. This MCP exposes their REST endpoints directly into your architecture. Your agents can invoke GPT-4o for complex JSON validation, route low-latency tasks to o3-mini, and generate vector embeddings natively. We abstract the streaming payloads and retry logic, ensuring your orchestration layer maintains deterministic control over token consumption.

GPT-4o & o3-mini text generation

DALL·E image creation & embeddings

Fine-tuning & Assistants API

Connect your agent

Fireworks AI MCP Server

Fireworks AI

Fastest hosted inference for open models

Together AI MCP Server

Together AI

Open-source model inference & fine-tuning

Replicate MCP Server

Replicate

Run any open-source model with one API call

Anyscale MCP Server

Anyscale

Ray-powered distributed model serving

Anthropic MCP Server

02 MCP Server

Anthropic MCP Server

When reasoning actually matters. Claude dominates complex coding tasks and 200K document parsing.

Anthropic provides superior performance in deep-context reasoning and multi-step tool execution. This MCP server connects your state machines to Claude's extended context window (200K+ tokens). Agents can pass massive codebase diffs or complex documentation arrays into the context boundary, extracting highly validated tool-use JSON payloads. The integration handles the strict XML-style prompt formatting required for optimal Claude inference.

Claude with 200K context & extended thinking

Tool use for agentic workflows

Vision analysis for images & documents

Connect your agent

NVIDIA AI MCP Server

NVIDIA AI

GPU-accelerated AI models & endpoints

NVIDIA NIM MCP Server

NVIDIA NIM

Optimized model containers for inference

NVIDIA API Catalog MCP Server

NVIDIA API Catalog

Browse & deploy 100+ GPU-optimized models

Baseten MCP Server

Baseten

Deploy custom ML models as API endpoints

Mistral AI MCP Server

03 MCP Server

Mistral AI MCP Server

Open weights that hit above their class. Mistral delivers serious efficiency and native JSON mode for tight workflows.

Mistral provides highly optimized, cost-efficient inference models via open weights. This MCP integrates your orchestration layer with Mistral Large for dense reasoning and Codestral for targeted syntax generation. It fully supports native JSON mode and strict function-calling schemas, allowing your systems to build high-volume, low-latency agent loops without overwhelming your API budget constraints.

Mistral Large, Small & Codestral models

Native function calling & JSON mode

Pixtral vision & multilingual excellence

Connect your agent

Helicone MCP Server

Helicone

LLM observability with cost & latency tracking

Langfuse MCP Server

Langfuse

Open-source LLM tracing & evaluation

Portkey MCP Server

Portkey

AI gateway with fallbacks & load balancing

LiteLLM MCP Server

LiteLLM

Unified LLM proxy for 100+ providers

Groq MCP Server

04 MCP Server

Groq MCP Server

LPU inference that runs faster than you can read. For agents where sub-100ms latency is a hard requirement.

Groq bypasses standard GPU bottlenecks using Language Processing Units (LPUs) to achieve sub-100ms inference latency. This MCP routes your API payloads through their hardware layer, accelerating open models like Llama and Mixtral. This integration is critical for synchronous agent workflows where real-time voice streaming or immediate UI state generation depends on deterministic, low-latency execution paths.

LPU-powered inference at 18x GPU speed

Llama, Mixtral & Gemma model access

OpenAI-compatible API with streaming

Connect your agent

Weights & Biases

ML experiment tracking & model registry

Arize AI MCP Server

Arize AI

ML & LLM observability platform

MLflow MCP Server

MLflow

Open-source ML lifecycle management

Braintrust MCP Server

Braintrust

AI evaluation & prompt engineering platform

Cohere MCP Server

05 MCP Server

Cohere MCP Server

Enterprise RAG done right. Command R+ and Rerank exist specifically to ground your agents in your actual data.

Cohere is engineered specifically for Retrieval-Augmented Generation (RAG) architectures. This MCP connects your agents to the Command R+ inference model and the Rerank API. The system enforces strict source grounding, preventing context hallucination by structurally forcing the model to cite injected vector chunks. We implement their multilingual endpoints to ensure your enterprise agents operate deterministically across vast internal knowledge graphs.

Command R+ with built-in RAG grounding

Multilingual embeddings in 100+ languages

Rerank for search result optimization

Connect your agent

AI21 Studio MCP Server

AI21 Studio

Jamba models for enterprise NLP tasks

Clarifai MCP Server

Clarifai

Computer vision & multimodal AI platform

New Relic AI MCP Server

New Relic AI

LLM observability inside New Relic APM

Datadog AI MCP Server

Datadog AI

LLM monitoring with Datadog APM integration

LLMs. Inference. Embeddings. Observability. Ready for AI Agents.

Free to start. Connect these servers to your AI agents in seconds -- no infrastructure to set up, no code to write.

Hosting, security, updates, and uptime -- all on us. You just connect and use.

Create Free Account Explore All Categories