Kindex
Give Claude Code or Cursor a durable knowledge graph across sessions with tiered context and inherited .kin project memory.
Overview
Kindex is a Build-phase MCP server that exposes a persistent knowledge graph with context tiers and .kin inheritance for AI workflows.
What is this MCP server?
- Persistent knowledge graph tuned for AI coding workflows
- Context tiers so agents pull the right depth without blowing the window
- .kin inheritance for project-scoped memory that travels with repos
- stdio MCP package on PyPI (kindex 0.4.1) with pipx runtime hint
- Documented at jmcentire.github.io/kindex for setup and concepts
- Server schema version 0.4.1 on PyPI identifier kindex
- Single stdio transport package in server manifest
- Hosted docs at jmcentire.github.io/kindex
Community signal: 15 GitHub stars.
What problem does it solve?
Agents forget project context between sessions and drown you in re-prompting unless memory is structured and reusable.
Who is it for?
Indie devs running long agent sessions who want repo-scoped memory graphs instead of ad-hoc markdown dumps.
Skip if: Teams that only need ephemeral chat with no cross-session memory or who cannot run a local Python MCP via pipx.
What do I get? / Deliverables
After registration, your agent can read and write a scoped knowledge graph so decisions and entities persist with tiered context across sessions.
- Registered stdio MCP connection to the Kindex knowledge graph
- Project-scoped graph nodes and relationships agents can query across sessions
- Tiered context pulls aligned to workflow depth
Recommended MCP Servers
Journey fit
Knowledge-graph MCPs sit where builders wire agent memory and workflow context into the stack, before ship and operate monitoring. agent-tooling is the canonical shelf for persistent memory layers that agents call over stdio during product work.
How it compares
Structured knowledge-graph MCP, not a flat notes plugin or a hosted vector DB SaaS.
Common Questions / FAQ
Who is Kindex for?
Solo builders and small teams using Claude Code, Cursor, or similar agents who need durable, project-scoped knowledge that survives session boundaries.
When should I use Kindex?
Use it while building agent-heavy products or multi-repo workflows where you want tiered context and .kin inheritance instead of repeating architecture in every prompt.
How do I add Kindex to my agent?
Install the PyPI kindex package (pipx is the documented runtime hint), add a stdio MCP entry pointing at the kindex server, and restart your agent host so graph tools are available.