proxy

A transparent MCP proxy and terminal UI for watching real client-to-server tool calls, inspecting frames, replaying captured calls, and debugging live JSON-RPC traffic.

Open Bias is an open-source reliability harness for LLM apps and agents that enforces rules at runtime. It sits between your app and an LLM provider, using RULES.md policies to trace, block, or fix off-policy behavior.

Phinq is an open-source runtime governance layer for AI agents that intercepts tool calls, classifies risk, pauses irreversible actions for human approval, and records decisions in a tamper-evident audit log.

A lightweight API key proxy for self-hosted llama.cpp servers with per-user token usage tracking, rate limiting, and multi-server routing.

LLMCap is a proxy for LLM API calls that hard-stops requests when a dollar cap is reached, with support for major providers and companion tools for VS Code, CLI, and Windows.

A local proxy that inspects LLM requests and attachments for secrets before they reach the model, with redaction, rejection modes, and local attachment decoding/OCR.

A non-destructive compression gateway for AI coding agents that sits between the agent and the LLM API to reduce token usage while keeping original content recoverable on demand.

A transparent SQL wire-protocol proxy that blocks destructive queries before they reach PostgreSQL or MySQL, with no application code changes required.

An open-source tool that screens AI agent tool calls and tool results before they are used. It works as an MCP proxy plus hooks adapter, with policy-based decisions, shadow mode, and local logging.

A portal layer that unifies multiple MCP servers behind one endpoint so AI clients can connect once and use three meta-tools instead of dozens or hundreds of exposed tools.

Open-source Go control-plane proxy for AI agents that enforces identity, policy, budgets, and audit for MCP and other upstreams, with anomaly detection that can auto-block compromised agents.

An AI API gateway/proxy in Go that sits between coding clients and upstream LLM providers, with request routing, secret redaction, context handling, MCP tool integration, and transparent SSE streaming.

Open-source transparent proxy for coding agents that preserves full conversation history on disk while selectively deciding what the model reads. Supports OpenAI-compatible, Anthropic Messages, and OpenAI Responses APIs, with setup docs, engine-specific configs, and benchmark claims for long-horizon agent sessions.

A local proxy and CLI for switching and monitoring Codex and Claude Code accounts, with combined token-throughput analytics and rate-limit awareness.

Open-source control plane and proxy for managing multiple MCP servers behind one endpoint, with routing, policy, auth, logging, and rate limiting.

A lightweight Flask-based LLM proxy that forwards Ollama, OpenAI, and llama.cpp-style requests to NVIDIA’s OpenAI-compatible API, with streaming support, retries, caching, auth, and usage stats.

A local Anthropic-compatible proxy that lets Claude Code work through ChatGPT, Kimi, Grok, or Cursor subscriptions, with provider routing, diagnostics, and model mapping.

Local-first record, inspect, and replay tool for debugging AI agents. Orchid captures LLM calls and tool traffic through a proxy, exposes traces in a web UI and MCP tools, and supports offline replay for deterministic testing.

A loopback HTTPS proxy that swaps placeholders for real API keys at request time, keeping secrets out of the agent process. It uses Bitwarden Secrets Manager and supports scoped bindings for hosts, methods, and paths.

Open-source MCP proxy that blocks agent send/share actions when the destination first appears in untrusted tool output. It uses deterministic rules rather than an LLM and logs every allow/deny decision.

A research project and proxy for Claude Code that chooses per-step reasoning effort with Jev while preserving the prompt cache. The repo includes benchmark results, a real-session measurement, and notes on how the mechanism works.

Open-source resilience and debugging proxy for autonomous AI agents. It checkpoints LLM and tool calls, supports cache recovery, human-in-the-loop approval, multi-model fallback, and exports trajectories for fine-tuning.

ContextSpy is an LLM context profiler that intercepts agent traffic, breaks down token usage, and visualizes how prompts change across requests and sessions.

A local proxy that lets Claude Code and other Anthropic API clients use Devin CLI’s model catalog and backend, translating Anthropic Messages API requests into Devin’s Connect-RPC format and streaming responses back.