Best Token AI Skills & MCP Servers

Token-efficient, markdown-first Notion MCP server — agents write markdown, never touch raw Notion JSON

MCP server for BYTE Library — per-byte USDC data feeds for AI agents on Arbitrum. Subscribe to first-party feeds or pay-per-call via x402; every settlement carries an EIP-712 PayloadAttestation receipt. No API keys, no token.

Model Context Protocol server for swatchbook — exposes a DTCG project's tokens, axes, and diagnostics to AI agents without running Storybook.

Model-agnostic code memory MCP server. Budget-aware graph retrieval for AI agents — sub-millisecond queries, token budgeting, no embeddings, no API keys. Built on CodeGraphContext for 155-language tree-sitter indexing.

ContextStream MCP server - v0.4.x with consolidated domain tools (~11 tools, ~75% token reduction). Code context, memory, search, and AI tools.

MCP server for token-aware code search and safe, atomic multi-file edits — built to cut Claude Code token usage on large codebases without giving up correctness.

Local MCP server for PixelAgent — apply diffs, resolve elements, read design tokens

Model Context Protocol server for Spectrum design data including tokens, schemas, and component anatomy

MCP server library for the Printr API - enables AI agents to launch tokens across EVM chains and Solana

MCP server that gives Claude Code persistent memory of your codebase, backed by Gemini's 2M-token context. Includes eager `ask`/`code` for cached repo Q&A and `ask_agentic` for agentic file access on repos that exceed the model's input-token limit.

LLM-native image optimization MCP server. Reduces vision model token consumption by snapping to tile boundaries.

Session metrics tracking — token usage, costs, duration, and per-tool statistics.

Local-first dashboard + MCP server for Claude Code and Codex token usage. MCP-aware, multi-vendor, MIT.

AHTML agent SDK — fetch snapshots from any AHTML-emitting site, cache by ETag, run actions with dry-run support, and measure token cost using the industry-standard tokenizers.

Token budget management — estimate and optimize token usage for AI agents.

Composite brick — orchestrates smartread, cache, compress, tokenbudget, and overview to deliver optimal context within a token budget.

YouTube MCP server for token-optimized, structured data using the YouTube Data API v3.

Pack files into a compressed context bundle — reduce token usage by extracting only signatures, maps, or full content.

Unified AI coding agent with MCP integration for Claude, VS Code, and Kiro IDE. Features persistent knowledge graph via OverdriveDB, token compression (60-80%), multi-model switching, and shared/independent project modes.

Track and report token savings — compare FocusMCP usage vs naive full-read, trend over time, ROI.

MCP server for Tokenmaxxen — collective episodic memory for AI coding agents. Deposit, retrieve, and comment on session learnings (gotchas, fixes, prompts, configs) over hybrid keyword+semantic search. Agents can post corrections when a learning is wrong

MCP server that reduces agent token usage by 90% with intelligent context compression, task checkpoint persistence, and workflow-aware agent guidance.

Focus your AI agents on what matters. 68+ bricks, 1 MCP server, modular context — from 200k to 2k tokens. Works with Claude Code, Cursor, Codex.

Universal MCP server for the Avito API — 145 tools across 18 domains, safe-by-default with dry-run, idempotency, structured errors, MCP resources/prompts and cross-process token lock. Install via npx, plug into any MCP-compatible client.