freellmpool, developed by 0xzr, is an open-source Model Context Protocol (MCP) server and API gateway that aggregates free tiers of dozens of Large Language Models into a single local endpoint. The tool pools multiple hosted providers, exposes an OpenAI-compatible interface, and offers routing, usage tracking, and keyless quickstart paths for immediate testing. Targeted at developers and AI-coding users, it reduces the effort of juggling many provider SDKs and keys during agent development.

Aggregates many hosted models and exposes MCP-native tooling for agents

freellmpool operates as an MCP server that catalogs over 24 providers and more than 400 chat models while exposing specialized tools such as 'ask', 'battle', and 'recipe' for AI clients. The tool includes shims for Codex-style and Anthropic-style messaging and lists compatibility with MCP-compliant clients, for example Claude Desktop and Cursor.

• Provider catalog: 24+ providers
• Model catalog: 400+ chat models
• Agent targets: coding agents like Claude Code, Cursor

Routing and cooldown logic preserve request flow but depend on external availability

Automatic failover and routing redirect requests when a provider returns rate-limit errors or goes offline, and the tool applies a smart cooldown that benches providers that hit limits to avoid repeated 429 errors. Usage tracking monitors per-day consumption so load is distributed; continuity therefore improves relative to single-provider setups, while actual uptime and throughput remain tied to the pooled free-tier services.

Low-setup integration for experiments, with practical limits for sustained workloads

Installation and integration target rapid experimentation: the tool installs via Python (pip), supports CLI use across desktop platforms, and provides an OpenAI/Anthropic-compatible proxy so existing software accepts the local endpoint without code changes. Keyless quickstart gives immediate access to certain providers, and users can add their own keys for higher quotas. Because it aggregates free tiers, the tool suits prototyping and agent testing rather than guaranteed, high-volume production runs.

Practical choice for experimentation, less suited to guaranteed production throughput

freellmpool is a practical option for developers and AI-coding users who need fast access to many hosted models through a single local endpoint; model behavior and availability reflect upstream providers, so expect variability tied to those free tiers. Treat outputs as upstream-model responses and supplement with provider keys if sustained capacity or stricter uptime guarantees are required.