MCP servers that work with local models: tested agent tools and MCP servers
As of 2026-10-22, TestedMCP ran 4 servers on small local models (qwen2.5-1.5b-instruct-q4km, qwen2.5-3b-instruct-q4km; CPU-only, 3 runs per server and model). 4 of 4 completed at least one task. The other 101 maintained servers are not yet tested on local models. We haven't run any 8B model yet.
Context cost: tiktoken cl100k_base over compact JSON of name+description+inputSchema (approximate). "Small-model friendly" is a heuristic from the real tool schemas (see methodology), not a model result. The 0-100 score is our standard score and doesn't measure local-model readiness.
| Server | Score | Tools | Small-model friendly | Schema context cost | Small-model runs |
|---|---|---|---|---|---|
| Sequential Thinking (reference) | 93 | 1 | medium | ~861 tokens (11% of 8k, 3% of 32k) | Not yet tested |
| Context7 (Upstash) | 100 | 2 | medium | ~1,074 tokens (13% of 8k, 3% of 32k) | Not yet tested |
| Exa | 93 | 2 | medium | ~414 tokens (5% of 8k, 1% of 32k) | Not yet tested |
| Tavily | 100 | 5 | high | ~1,612 tokens (20% of 8k, 5% of 32k) | Not yet tested |
| Memory / knowledge graph (reference) | 93 | 9 | high | ~925 tokens (11% of 8k, 3% of 32k) | qwen2.5-3b-instruct-q4km: 100% tasks completed, 100% correct tool, 3 runs; qwen2.5-1.5b-instruct-q4km: 67% tasks completed, 100% correct tool, 3 runs |
| Filesystem (reference) | 93 | 14 | high | ~1,666 tokens (20% of 8k, 5% of 32k) | qwen2.5-3b-instruct-q4km: 100% tasks completed, 100% correct tool, 3 runs; qwen2.5-1.5b-instruct-q4km: 100% tasks completed, 100% correct tool, 3 runs |
| Playwright MCP (Microsoft) | 100 | 25 | medium | ~3,688 tokens (45% of 8k, 11% of 32k) | Not yet tested |