MCP security scanner - detect prompt injection in tool descriptions

MCP proxy that compresses prose fields (tool descriptions, etc.) using caveman rules. Same accuracy, fewer context tokens.

OpenCode plugin that reduces token usage by up to 33% — compresses tool descriptions, compacts read output, adds line-range edit support

Shared client infrastructure for Vultisig CLI and MCP — auth, config, tool descriptions

MCP server that teaches any agent the AIDE methodology through tool descriptions and progressive disclosure tooling

Slim OpenCode system prompt and built-in tool descriptions to reduce per-request token overhead. Plugs into the tool.definition and experimental.chat.system.transform hooks.

Measure how well AI agents actually use your MCP server's tools — then rewrite the tool descriptions and prove the improvement with before/after data.

MCP server that helps AI agents stay reliable by scoring tool descriptions, estimating token costs, simulating tool choice, and generating tests.

MCP server for SearXNG with engine/category targeting, multi-page fanout, and dynamically-generated tool descriptions tailored to the instance's actual enabled engine pool.

Pi extension — auto-translate all tool descriptions to your locale via LLM

MCP security scanner - detect prompt injection in tool descriptions

MCP server for Rakuten Web Service: 28 read-only tools across Ichiba, Books, Travel, Recipe, Kobo, and GORA. Bilingual EN/JA tool descriptions.

On-demand MCP tool activation for OpenCode — saves ~80% tokens by shrinking MCP tool descriptions and routing via keyword matching

Load YAML files containing LLM tool descriptions for MCP, Anthropic/OpenAI tools, Zod.

MCP server agent-usability scorer — scores and auto-fixes tool descriptions

OpenCode plugin that reduces repeated OMO task output, safely distills reconstructable background results, and compresses built-in tool descriptions.

A transparent MCP security proxy: inspects tool descriptions for injection, pins tool definitions to detect rug-pulls, enforces a default-deny capability policy, and writes a tamper-evident audit log.

MCP server for emulators that speak PINE — first-class PCSX2 (PS2) and RPCS3 (PS3), with target-aware tool descriptions, memory r/w, and savestate control

Dynamic tool loading for opencode — Kimi K3 tool-search pattern, adapted to opencode's plugin runtime. Collapses noisy tool descriptions into a searchable catalog so the LLM only sees relevant tools per turn.

Local static scanner for MCP setup, prompts, tool descriptions, and repo manifests.