Clerk, from Vulcanshen, is a local Model Context Protocol server that preserves session context for AI-assisted coding. It records developer activity and uses configured AI backends to generate condensed summaries and structured progress reports. The tool provides explicit session markers, searchable history, and automated daily and weekly reporting to reduce manual documentation. It suits engineers who want context recovery without relying on cloud-based tracking tools.

Captures work into explicit sessions and searchable logs

The tool relies on explicit start and end markers to define discrete work blocks and records metadata and messages into a local SQLite store. That storage model supports a searchable history and the generation of session summaries, so past decisions and timestamps remain retrievable. Context recovery hooks allow MCP-compatible assistants to reference stored session state and continue a paused workflow without re-supplying earlier chat content.

Summary usefulness depends on the configured AI provider

The tool generates automated session summaries using a configured processing model; it is provider-agnostic and supports OpenAI and Anthropic backends via environment variables. Because summaries are produced by the selected remote model, their factual accuracy and tone reflect that provider's outputs. Users should review and edit generated notes before treating them as definitive project documentation.

Installation and integration expect developer familiarity

The tool is a Node.js application that runs on Windows, macOS, and Linux and requires an MCP host such as Claude Desktop plus a Node.js runtime. Installation paths include npm or npx, or cloning the repository and building with TypeScript. Configuration uses environment variables to select AI backends, so integrating Clerk into an existing coding workflow assumes comfort editing env vars and running a local server.

Local-first storage reduces automatic uploads while external models remain optional

The tool stores session logs and metadata in a local SQLite database and does not automatically upload that data to external servers. At the same time, summarization can be routed to externally hosted models when users configure backends. Consider these behaviors when handling sensitive code:

• Logs kept locally: SQLite files remain on the developer's machine
• Summaries via configured providers: network calls occur if OpenAI or Anthropic are selected

Practical choice for MCP workflows with an on-device memory preference

The tool suits developers using MCP hosts who want persistent, on-device session memory tied to their workstation. Expect to treat generated summaries as editable drafts and to plan for local Node.js hosting and environment-variable configuration. For teams that need a memory layer anchored to developer machines rather than a cloud tracker, the tool offers a pragmatic way to keep multi-session project history accessible.