MCP context layer for coding agents

Contextful

Highly efficient context management for Codex, Claude Code, Cursor, Windsurf, GitHub Copilot, VS Code, Cline, Roo Code, Continue, and Zed.

Install with npx View source

context_pack("where are MCP tools registered", budget: 600)

7 cited hits across 2 files
saved roughly 18 tool calls
returned a token-budgeted evidence pack
ready for follow-up trace, impact, and memory queries

The Layer Agents Need at Runtime

Solid Search Engine

Classifies each query and combines FTS, symbols, docs, graph expansion, and deterministic reranking.

Efficient Context Storage

Stores files, chunks, symbols, graph nodes, graph edges, fingerprints, packs, and memory in SQLite.

Memory Ledger

Agents can store lessons only when they cite valid evidence from files, symbols, or prior context packs.

Install

npx @inferensys/contextful index --workspace .
npx @inferensys/contextful query "where is auth handled" --workspace .

codex mcp add contextful -- \
  npx -y @inferensys/contextful server

Use context_pack as the primary agent tool. It returns ranked, cited, token-budgeted evidence instead of forcing dozens of random file reads.