Code-only MCP servers — agent-tools.cloud

MCP servers that ship as code — a repository or a package you install and run locally (usually over stdio). They have no hosted endpoint, so we cannot probe or grade them: instead each one shows its repository, install command, license, stars and when it was last updated.

Collected from GitHub repositories and npm, PyPI, Cargo, Go, Maven, NuGet and RubyGems packages tagged as MCP servers, the official MCP Registry, Docker's MCP catalog, ModelScope, Baidu MCP World, Cline and other public directories; one entry per repository. A server that also offers a hosted endpoint is listed under Remote instead. These entries are not counted in our reachability, verification or readiness statistics.

token-shrink

github.com/ajdev0/token-shrink
active

Local-first token reduction engine and MCP server: gives AI coding agents dependency skeletons (types, signatures, exports) instead of full files, so they read far fewer tokens.
astermind-mcp

github.com/AsterMindAI/astermind-mcp
active

AsterMind MCP server - on-device RAG token reduction (stdio, 10 tools)

MIT ★ 1
indexio

github.com/IOServicesLabs/indexio
active

Code Search and MCP server for AI coding agents. Instant Token Reduction.

Apache-2.0 ★ 9
jidra

github.com/akhilsinghcodes/jidra
active

Framework-aware code intelligence for LLM agents. Indexes Java, Python, TypeScript, Go, Scala into a verified call graph via tree-sitter + Rust. MCP server, Haiku sub-agent, and git hooks included. 68–95% agent token reduction.

MIT ★ 1
atr-mcp

github.com/jacobbhatt12/atr
active

ATR: Automation for Token Reduction. An MCP proxy that hides tool lists and shrinks tool results so your AI uses fewer tokens.

pypi MIT
stump

github.com/hegner123/stump
updated this year

Token-efficient directory tree visualization MCP server in Zig - optimized for LLM consumption with 50%+ reduction vs standard tree output

MIT
shadow-web

github.com/ulinycoin/shadow-web
active

Local-first Python SDK for AI agents: Shadow DOM flatten, Action Map, SchemaSnap, form fill, security scan, MCP (64–97% token reduction)

MIT ★ 10
@fastmcp-me/mcp-sqlew

github.com/fastmcp-me/mcp-sqlew
stale

MCP server for efficient context sharing between Claude Code sub-agents with 96% token reduction via action-based tools

npm MIT
jev-codex-workbench

github.com/qualixar/jev-codex-workbench
active

Codex plugin and MCP server for TypeSafe Jev decisions and recoverable context reduction, with optional local Laya-MLX and token/time measurement tools.

NOASSERTION ★ 1
capsule

github.com/hakiyaka/capsule
active

Codex plugin MCP server (codex-plugin-mcp) for token efficiency, token reduction, context compression: exact-recoverable terminal, file, project, web, memory.

MIT ★ 4
lookacrawler

github.com/lucasmartins-ai/lookacrawler
active

Free, open-source, token-efficient local alternative to Firecrawl with native MCP Server for LLMs (>73% token reduction).

MIT ★ 5
recursive-thinking-mcp

github.com/parth3930/recursive-thinking-mcp
updated this year

Token-efficient recursive thinking MCP server for AI agents. Achieve production-ready solutions through iterative refinement with 70% token reduction. Perfect for Claude, AI problem-solving, and complex task automation.

npm MIT
mcprune

github.com/hamr0/mcprune
updated this year

MCP middleware that prunes Playwright accessibility snapshots for LLM agents. Zero ML, 75-95% token reduction, all refs preserved.

npm MIT
trimtok

github.com/Stxyu-p/trimtok
active

Universal, deterministic context compressor for AI coding agents & IDEs (Claude Code, Cursor, Antigravity, Codex, Gemini, Hermes) — 88.3% token reduction, lossless TOON v2 JSON tables, AST code skeletonization, and SQLite WAL + Zstd CCR (#c1) recovery over stdio MCP.

pypi MIT ★ 3
mcp-codebase-index

github.com/MikeRecognex/mcp-codebase-index
updated this year

17 MCP query tools for codebase navigation — functions, classes, imports, dependency graphs, change impact. Zero dependencies. 87% token reduction.

AGPL-3.0 ★ 63
uspto_enriched_citation_mcp

github.com/john-walkoe/uspto_enriched_citation_mcp
active

High-performance MCP server for USPTO Enriched Citation API v3 with AI-powered data extraction, token-saving context reduction, progressive disclosure workflows, and seamless cross-MCP integration

MIT ★ 7
@csc-protocol/token-saver
active

Autonomous AI agent tool and MCP server for semantic token compression and L402 edge solvency clearance.

npm UNLICENSED
mcp-token-optimizer

github.com/techdeveloper-org/mcp-token-optimizer
active

Cuts Claude Code and LLM token usage by 60-85% via AST navigation, deduplication, and context-budget optimization.
tokendamper

github.com/ojassug/tokendamper
active

Universal context optimization for AI coding assistants.

npm MPL-2.0
token-pilot

github.com/digital-threads/token-pilot
active

Save up to 80% tokens when AI reads code — MCP server for token-efficient code navigation, AST-aware structural reading instead of dumping full files into context window

npm MIT
json-to-toon-mcp-server

github.com/your-username/json-to-toon-mcp-server
updated this year

MCP server for converting JSON to TOON (Token-Oriented Object Notation) format - saves 30-60% on AI tokens

npm MIT
tokensaver

github.com/pozii/tokensaver
updated this year

MCP server that cuts AI agent token costs by up to 97% — compression, caching, pruning, web extraction

Apache-2.0 ★ 1
mcp-context-condenser

github.com/rafim-dev/mcp-context-condenser
active

Token-slimming AST code outliner, log compressor & context budget analyzer for AI coding agents (Cursor, Claude, Cline, Antigravity). Slashes token usage & LLM API bills up to 85%.

MIT
five-mcp

github.com/kiro0x/five-mcp
updated this year

160,000 deductively-derived JSON constraints that eliminate LLM persona drift. Measured: 0 breaks in a 120-turn pressure test.

pypi MIT
pcp-engine

github.com/rishi-banerjee1/prompt-control-plane
updated this year

PCP — the prompt quality engine. Deterministic scoring, compilation, cost estimation & governance for AI prompts. Zero LLM calls.

npm Elastic-2.0
cctx-optimizer

github.com/anil7948/cctx
updated this year

Reduce Claude Code token usage by 70-90% with cross-session memory and four-layer optimization: semantic codebase indexing, PostToolUse tool compression, turn summarization, and session consolidation. MCP server powered by Ollama — zero API calls, runs en

npm MIT
@blanklogic/refinery-mcp
updated this year

Clean context before AI sees it — BlankLogic Refinery MCP for Cursor and Claude Code (local, no upload).

npm UNLICENSED
claude-prompt-optimizer-mcp

github.com/rishiatlan/prompt-optimizer-mcp
updated this year

The control plane for AI prompts — deterministic scoring, policy enforcement, audit & governance via MCP. Free tier included.

npm Elastic-2.0
latent-gate

github.com/kathanmodh259/latent-gate
active

Local-first vision-language pipeline inspired by VL-JEPA. Compress images, text, conversations, and RAG documents locally via Ollama before sending to any LLM API. Deterministic, fact-preserving token optimizer plus MCP server, FastAPI server and video processing.

pypi Proprietary
ai-coding-context-optimizer

github.com/ehk2509/ai-coding-context-optimizer
active

Token and context optimization for AI coding agents with smart compression, retrieval, recovery, routing, and MCP integrations.

MIT ★ 1
cctx

github.com/anil7948/cctx
updated this year

Reduce Claude Code token usage by 70-90% using a free local LLM (Ollama). MCP server + Stop hook with codebase indexing, tool output compression, and turn summarization.

MIT ★ 3
slm-gate

github.com/zenithfoundry/slm-gate
active

Decoupled local-SLM pre-processing and routing layer for MCP and cloud LLMs. Intercepts, distills, and routes prompts locally with Ollama to cut token costs and protect monthly quota.

npm MIT ★ 3
latent-gate

github.com/KathanModh259/latent-gate
active

MCP server that shrinks logs, JSON, docs and web pages before your AI reads them. Offline, deterministic, ~3ms, no API key. 66% fewer tokens on a mixed dev corpus, 90%+ on logs.

NOASSERTION ★ 25