memex-mcp
Persistent memory for AI coding agents. Builds a temporal knowledge graph of your codebase from commits and source files, then serves it thr
Open source Repository Open in the app JSON README (API)
About
Persistent memory for AI coding agents. Builds a temporal knowledge graph of your codebase from commits and source files, then serves it through 14 MCP tools so every Claude Code session starts knowing your architecture, recent decisions, and open problems — no more cold starts or context pasting. Bitemporal facts (every edge knows when it was true), two-regime confidence decay (unvalidated facts cross stale at exactly 30 days), hierarchical Leiden clusters keep get_project_context under 1500 tokens for any repo size, and human-in-the-loop validation. Backed by Neo4j + Graphiti + Gemini Flash. Also publishes to PyPI and npm.
Details
- Kind
- Plugins
- Topic
- AI, RAG & memory
- Publisher
- stifler7
- Origin
- marketplace
- Category
- ferramentas
- Last push
- 2026-09-07T01:48:13Z
- Repository state
- ativo
- Language
- Batchfile
- License
- MIT
- Added
- 2026-08-30 01:48:58
- Updated
- 2026-08-30 01:48:58
- Origin id
stifler7/claude-plugins/memex-mcp
README
# STiFLeR's Claude Code Plugins [](https://github.com/STiFLeR7/claude-plugins/stargazers) [](https://www.npmjs.com/package/stifler-memex-mcp) [](https://pypistats.org/packages/memex-mcp) [](LICENSE) A small, opinionated marketplace of Claude Code plugins focused on durable engineering context and grounded agent workflows. ``` /plugin marketplace add STiFLeR7/claude-plugins /plugin install memex-mcp@stifler-marketplace /plugin install prompt-forge@stifler-marketplace ``` ## Plugins ### 🧠 [memex-mcp](./memex-mcp) Trusted engineering context for AI coding agents. Builds a temporal knowledge graph of your repository and exposes bounded, provenance-aware context through 14 MCP tools — `get_project_context`, `record_decision`, `predict_impact`, `explain_change`, and more. Claude Code can explicitly recover your architecture, decisions, open problems, and code relationships without treating the plugin as a replacement for Git or Claude's own session memory. - Bitemporal facts with expiry and supersession - Confidence decay, validation, corroboration, and stale-context review - Bounded `ContextPacket` projections with provenance and selection reasons - Human-in-the-loop validation via `memex review` - Explicit, governed writes for decisions, problems, resolutions, and invalidation Backed by Neo4j + Graphiti; optional LLM backends support synthesis and explanation. MIT licensed. → [github.com/STiFLeR7/memex](https://github.com/STiFLeR7/memex) ### ✍️ [prompt-forge](./prompt-forge) Advanced prompt refinement skill. Extracts your *real* intent, investigates your codebase and ecosystem, applies 9 perspective lenses (Security, QA, UX, Performance, …), and produces a grounded prompt for your execution tool — without ever executing the task itself. Trigger phrases: *"help me prompt"*, *"improve my prompt"*, *"how to write a prompt for…"*. The skill investigates, asks sharp clarifying questions, then hands you a refined prompt to paste into your next agent. MIT licensed. → [github.com/STiFLeR7/prompt-forge](https://github.com/STiFLeR7/prompt-forge) ## Local development ```bash # Clone and link as a local marketplace for testing git clone https://github.com/STiFLeR7/claude-plugins /plugin marketplace add /absolute/path/to/claude-plugins ``` ## Contributing Plugins added to this marketplace should be: - Single-purpose and well-tested - Documented with clear trigger conditions - MIT (or compatible) licensed - Maintained by an active author Open an issue with a proposal before submitting a PR. ## License MIT — see individual plugin directories for their respective licenses.