"Scratch" matching MCP servers:

An MCP server that exposes the markdown issue tracker under .scratch/ as a queryable graph, letting agents see open, blocked, or takeable issues without re-parsing prose.

A codebase-memory MCP server for AI coding agents with hybrid semantic+BM25 code search, structural outlines, a work journal, and gaze-and-think deep reasoning over a local LanceDB database.

MCP server providing persistent memory and context for AI tools, including semantic memory, knowledge graph, and session history to avoid starting from scratch in every conversation.

Enables autonomous agents to snapshot execution-tree and context states, evaluate branch viability, and roll back to previous checkpoints when an exploration path fails without restarting from scratch.

Enables shared scratchpad functionality for Claude Code sub-agents with full-text search, workflow organization, and intelligent Chinese text tokenization. Supports seamless context sharing and collaboration across workflows with project scope isolation.

Enables long-running autonomous agents to snapshot environment and context states, evaluate the viability of competing reasoning branches, and roll back to an earlier checkpoint when an exploration path fails without restarting from scratch. Exposes this tree-of-thoughts state management over the Model Context Protocol for use in Claude Desktop, Cursor, and custom agentic frameworks.

scratchpad-mcp is an MCP server that gives AI agents persistent, token-efficient storage. Eight tools versioned file writes, structured line-diffs since a prior version, append-only logs with cursor pagination, on-demand summaries with per-version caching, and per-agent quotas. The point is to stop agents from re-reading files they've already seen and re-summarizing documents.

Compiles locally captured screen activity into deterministic, evidence-backed records and executable workflows that agents can consume through MCP tools. Enables agents to fetch recent context, day summaries, click-by-click replay steps, repetitive-workflow patterns, and communication surfaces so routine tasks can be repeated instead of re-derived from scratch.

Gives Claude Code sessions a durable, shared memory and work ledger backed by a local SQLite database, exposing the brain_* tools so agents can create and lock goals with verifiable success criteria, plan and track work units, record evidence and failures, and search decisions, learnings and project knowledge. Because the tool descriptions carry the operating rules, every session in every project reads and writes the same accumulating context instead of starting from scratch.

Give your coding agent the repo context it needs before it starts searching. Madar builds a local graph of your TypeScript or Node.js repository and turns the current question into a small, task-aware context pack. Claude Code, Codex, Cursor, Copilot, Gemini, Aider, and OpenCode can start from relevant files, symbols, snippets, and relationships instead of rediscovering the repository from scratch