"How to create the most intelligent agent" matching MCP servers:

Scaffold and audit Starter Series projects — MCP server, Claude Code skill, and CLI in one package.

Enables ordinary LLMs to be augmented fully offline — with no API keys or external services — through model-aware guardrails, exemplar retrieval, structured self-critique, compliance scoring, and an allowlisted verification runner. Also provides hybrid knowledge-base search, decaying long-term memory, and persistent plan/task management across 54 MCP tools usable from any MCP client such as OpenCode, Claude Desktop, Cursor, or VS Code.

A standalone stdio MCP server that exposes 13 resources, 3 prompts and 7 tools for assembling interactive, replayable HTML interfaces such as calculators, charts, maps, flowcharts and small apps. It returns the generated HTML document for the calling client to decide whether and how to display.

Enables AI agents to control a Live2D desktop pet's expressions and actions via MCP protocol.

An MCP server that generates and edits images using OpenAI's GPT Image model, allowing users to create images from text descriptions and edit existing images through natural language.

Local MCP server that connects VS Code agents like Cline to OpenAI and provides workspace file tools (read, write, list) with restricted file access.

Enables AI agents like Claude to interact with Figma designs through 50+ tools for creating, styling, and manipulating design elements, components, and variables via a WebSocket relay and Figma plugin.

Enables an agent to submit a task brief and receive a short, validated list of skills from its full catalogue that are worth reading before starting.

It enables AI agents to call a small set of reliable, permission-aware tools generated from an existing website through the Model Context Protocol, without rebuilding the site.

MCP server for managing i18n translation files — gives your AI agent full control over your app's translations without dumping entire locale files into context.

Enables structured learning with a verified loop: define goals as observable claims, learn through teach-lab-test-gate per claim, and get independently graded by an adversarial examiner to ensure genuine progress.

A minimal MCP server using FastMCP to demonstrate tool creation, including a get_weather example.

Give any LLM agent a real Android or iPhone. 62 MCP tools: tap, swipe, type, screenshot, screen-tree reading, app launch, camera, TTS, crash reports, batched execution. Android via ADB, iPhone via WebDriverAgent, on-device inference, Docker+KVM emulators. Works with Claude Code, Cursor, LangChain, LlamaIndex, and any MCP client. MIT.

Enables a Codex orchestrator to obtain a deterministic, local recommendation for a model and reasoning effort for a delegated coding task, without inference calls or API keys.

Skill librarian for coding agents: describe a task and get the few matching skills from a large local skill collection, then report whether they worked. Runs locally with Ollama and SQLite.

Enables AI agents to directly manipulate Jupyter Notebook cells and structure through a secure, well-defined API, overcoming Cursor's AI agent mode limitations.

A local MCP daemon that turns an agentic coding client into a bug bounty operator, with 103 tools for offensive security testing including MITM proxy, traffic analysis, and OOB callbacks.

Enables MCP-capable agents to locate game save files, identify their engines, compute unlock plans for gallery/recollection flags, diff saves, and safely dry-run, apply, verify, or roll back reversible byte patches with automatic backups.

Enables agents to convert Markdown into WhatsApp-compatible formatting, including styled text, lists, and tables that fit a phone's monospace width.

Enables MCP clients to turn Make blueprints and Zapier exports into n8n workflow files, with a per-step report of what converted, what to review, and what must be finished by hand. Also lists zaps, explains how individual steps are handled, checks workflow files for broken links or placeholders, and can upload the result to an n8n instance as an inactive workflow.

AI-powered MCP server for file renaming, organization, and PDF splitting using natural language prompts and content analysis.

Routes one brief to the right image model across 60+ (gpt-image-1.5, Ideogram 3, Recraft V4, Flux), validates the output, and fans out to iOS/Android/PWA/favicon/visionOS/Flutter bundles. Works without an API key via Pollinations, HF Inference, Stable Horde, or host-LLM inline SVG.

Enables turning requirements documents into RFCs, functional test scenarios, and a traceability matrix, with tools to parse, diff, and verify coverage.

Enables searching and retrieving documentation from crawled documentation sites as an MCP server, allowing coding agents to query real docs instead of relying on training data.

A lightweight MCP bridge that lets Claude Code delegate heavy tasks to the Antigravity CLI - saving context window and tokens for what matters.

A Model Context Protocol server that converts Markdown content to HTML format.

Enables an MCP-compatible AI client to author Blender Python, run Blender, import the resulting asset into Unity with materials and a prefab, place it in the active scene, verify the import, and revise the same asset using stable IDs.

This MCP server lets AI agents connect to Figma to read, analyze, create, and modify designs, including elements, styles, assets, and code generation. It works with free Figma accounts and supports parallel multi-agent use across MCP-compatible clients.

A local stdio MCP server that enables researchers to practice explaining their work outside a conference setting through a guided, one-question-at-a-time interview about their research, its audience, its consequences, and a concrete next step. It stores the saved answers in a local SQLite draft and returns an approximate rubric-based grade, without calling any model or requiring an API key.

Enables AI coding agents to act as a database administrator by decomposing query plans, inspecting schema and index coverage, generating index recommendations, and dry-running DDL migrations inside a savepoint that always rolls back while flagging risky patterns.

Enables MCP clients to read bounded workspace files, validate exact replacement plans for the VS Code extension, and check recent typing progress without modifying source or running shell commands.

Enables MCP clients to interact with A2A agents through four ordinary tools, translating agent discovery, messaging, task reading, and cancellation between the A2A and MCP protocols, with support for the tasks extension.

Deterministic Markdown-to-Word (.docx) publishing compiler with native OMML equations, three-line tables, CSL citations, and quality gates for AI agents.

MCP server and Figma plugin that let Claude build real design systems in Figma Design: variables, components, styles, batch binding and whole screens in one call. Runs locally through the Figma plugin API, with no usage limits and no paid seat.

An MCP server that manages your conversation context automatically, pruning dead branches and keeping an audit trail of every cut.

Enables an agent to turn a paper's draft into editable Figma frames for publication-ready figures, checking every number against the paper and verifying the exported PDF.

Enables forwarding Douyin or Bilibili video shares to Kimi for AI analysis and automatically writing the resulting Chinese notes into an Obsidian vault.

Enables an MCP client to retrieve explicitly allowlisted Markdown context, create and read local task contracts, write local handoffs, and return locally configured routing metadata without invoking a model, executing commands, or writing to the vault.

Enables Claude to translate natural-language business questions into safe, read-only SQL queries by exposing tools for listing tables, inspecting schemas, and executing row-capped SELECT statements against PostgreSQL.

Exposes 11 agent-native QA tools over stdio so coding agents can discover a project, analyze risk, select the smallest relevant test set, generate and dry-run tests, triage failures, propose tiered heals, analyze flakes, generate quality reports, and evaluate release readiness. Planning and healing tools stay read-only or propose-only, while execution defaults to dry-run, keeping agent actions safe and evidence-backed.