"author:0rkz/byte-mcp-server" matching MCP servers:

Furl folds repeated tool output into a hash-addressed marker, keeps the line that matters, and returns any original byte-exact the moment the agent asks for it. No summary. No guessing. Nothing thrown away.

Scans prompts for PII and masks or redacts sensitive data locally before sending to an LLM, supporting multiple anonymization modes.

Persistent AI memory for Claude Code, Cursor, GitHub Copilot & Windsurf — sessions, lessons learned, semantic search, and team brain. 38 MCP tools. Free tier, EU servers.

AI-powered academic paper search MCP server with relevance scoring, summarization, author search, and credit checking. Enables users to search papers naturally and get scored results.

This server enables decidable, hardware-attested semantic verification of AI outputs using on-chip ballistic walks, providing reproducible receipts that can be recomputed byte-for-byte.

Converts video and image files through the grey telecinema presets (NTSC-J seated black, 8-bit grey, 1-bit overcalc, and a 4x4 collage), producing silent, dither-free output suitable for AI models to ingest. Every conversion is logged with its full command line to an always-visible SQLite index, so byte and dimension changes between files can be audited at any time.

Hosted MCP server for structured knowledge extraction and entity enrichment, allowing users to author JSON schemas, enrich entities using multiple LLM providers, and benchmark model performance.

Enables MCP clients to access upstream MCP capabilities and OpenAI-compatible model routes through one invariant-preserving, policy-bounded endpoint with deterministic routing, leases, durable Tasks/Runs, and verifiable outcomes.

Lossless context compression for LLMs, packing text into 2x-8x fewer tokens with byte-exact reconstruction. Provides MCP tools to compress files/text and expand exact slices, verified by sha256.

Enables remote MCP access to tools for URL-to-markdown conversion, web and image search, academic search, embeddings, reranking, classification, deduplication, and PDF extraction.

Author verifiable eval records through a draft → review → revise → submit loop with server-enforced graders; compile to JSONL/CSV/Inspect/lm-eval via MCP.

Buying-intent judgement for AI agents: three judges decide whether the author of any post or message is a real buyer, and draft a reply you approve.

Enables users to coordinate planning and multi-model code execution by creating project-scoped chats, dispatching tasks to configured models, retrieving replies and status, and sending queue or steer follow-ups through its available tools.

A local MCP server that brings Chinese NLP capabilities to AI agents and IDEs. It provides four tools: Chinese word segmentation (jieba), pinyin conversion (pypinyin), TF-IDF keyword extraction, and custom sensitive-word detection (Aho-Corasick). All processing runs locally over stdio, with zero network requests and no built-in sensitive-word dictionary — users supply their own word lists. Install

Enables admins to author and manage persistent mock MCP servers and OpenAI-compatible endpoints through a browser UI and admin API, with REST mocks, shared datasets, and traffic logging for demos and training.

Shrinks logs, JSON, docs and web pages before your AI reads them: offline, deterministic, no API key. Folds repetitive log lines (12,037 to 127 tokens on a 400-line log) while code, URLs and quoted strings come back byte-for-byte. Also fetches URLs as clean text, counts tokens exactly, and has optional local Ollama tools for image and text compression.

Resolves what an MCP tool call would do by planning it: it merges inputs, substitutes static placeholders or asks a model to author commands and queries, checks them against guardrails, and returns the masked plan with targets plus a dispatch verdict. It executes nothing and connects to no database, gateway, or queue, so it can run anywhere and be tested without infrastructure.

Enables Claude Code to delegate tasks to free OpenRouter models as callable tools while keeping the Anthropic endpoint for the orchestrator, with optional sandboxed file and shell access.