Lcc

Lcc is a coding-agent context compiler that reduces token counts while keeping content byte for byte.

Lcc is a local context compiler for coding agents that compacts context without summarizing it away. It cuts what the model does not need while keeping user and assistant messages byte for byte. Removed tool outputs move to a local content-addressed store and stay recoverable via recall. It includes a Claude Code plugin, an MCP server, and a CLI for processing session files or documents. The tool runs locally on the CPU by default in plain Python without requiring an API key or network access.

Lcc at a glance

Best for: Software engineer
Runs on: Web, CLI, API, MCP
Pricing: Free
Released: October 2026
Category: AI coding tool
Website: github.com/lucasmartins-ai/lcc

Key features

Stashes removed outputs for local byte-level recovery
Runs locally on CPU in plain Python without a network
Provides a Claude Code plugin for session compaction
Includes an MCP server for stdio clients
Generates JSON audit reports for every pass
Supports multiple providers including specialist-v3 and mechanical
Provides a CLI to process session files and documents

Use it for

Compacting Claude Code session transcripts to save tokens
Recovering removed tool outputs during coding agent sessions
Analyzing token usage and input distribution in sessions
Running local context compression offline without an API key

Companions

Tools that articles about Lcc recommend using with it, each with its job.

LangGraph — Durable execution, checkpoints and human approval under create_agent.
LangSmith — Step-by-step traces and evals that show what an agent actually did
Pinecone — Managed retrieval for the fastest path to production.
Qdrant — Self-hosted vectors when cost or data residency matter.
Ollama — Fully self-hosted inference paired with Qdrant.
Langfuse — Self-hostable alternative to LangSmith.

Alternatives to Lcc

Compactly — Compacts sessions at natural breaks and shows turn costs and cache expiration.
Ctxfw — Reduces AI context token mass while preserving code structure and dependencies.
Lcu — Runs the original computer-use runtime from the ChatGPT app without sign-in.
claude-mem — It shares persistent, searchable agent memory across sessions and devices.
LangChain — Compose exactly the agent your use case needs from model, tools, prompt, and middleware.
Agent Archive — Hand off coding sessions between Claude Code, Codex, and Cursor across machines.
ECC — It provides a shared operating layer and security scanning across multiple coding agents.
Claude Context — It reduces token costs and provides deep context for large codebases.
jevgrep — Reduces coding agent costs by providing targeted, relevant repository context.
Lessloc — Reduces AI coding agent code output by up to 36 percent and costs by 26 percent.
LangChain4j — Integrates LLMs and vector stores directly into Java applications.
Context Mode — Saves up to 98 percent of context per session by keeping raw data out of the LLM.

All Lcc alternatives