context_pack
Return a ranked, cited, token-budgeted context pack so the agent can avoid broad file reads.
Give a coding agent one ranked, cited, budget-capped bundle of project context instead of forty file reads at the start of every session.
LABSCO SUMMARY
The context pack is the point: it comes back ranked, with citations, and sized to the budget you passed, so the cost of establishing context stops scaling with the size of the repository. The memory ledger is deliberately strict — a lesson has to carry evidence references, and when the evidence underneath it changes the memory is marked stale, which is what keeps stored notes from quietly going wrong. Nothing leaves the machine.
A local index over one workspace — SQLite with full-text search and a typed graph of files, symbols, modules and config — queried through a small tool surface, plus a memory ledger for lessons that outlive a session.
No account and no key — v1 is local-only: it calls no external embedding API, uploads no source, edits no files and installs nothing inside the workspace it indexes. Run it as a server with npx -y @inferensys/contextful server over stdio, from the npm package @inferensys/contextful, currently 0.1.0, which also installs the cxf and contextful binaries. The workspace has to be prepared first: cxf init --workspace <YOUR_WORKSPACE> then cxf index, with --watch or cxf daemon to keep it fresh as files change. State lives in a .contextful/ directory. The project describes itself as early alpha with breaking changes expected.
Follow the published steps
USE CASES
Ask for a context pack on the question and read the cited bundle rather than opening files at random.
Search across code, docs, symbols and stored context with the ranking doing the narrowing.
Pull the likely dependents of the symbol or file and the tests that cover them.
Trace the graph path between them along the edge types that matter.
Combine the current evidence with its git history in one answer.
Write the lesson with the evidence refs behind it, and recall it at the start of the next session.
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
context_packReturn a ranked, cited, token-budgeted context pack so the agent can avoid broad file reads.
impact_analysisFind likely dependents and tests for a file, symbol, or module.
recall_memorySearch the evidence-backed memory ledger for lessons that survived previous agent sessions.
search_codeSearch code, docs, symbols, and stored context with lexical, symbol, and graph-aware ranking.
trace_pathTrace graph relationships between files, symbols, modules, and config nodes.
why_changedExplain why a file or symbol may have changed by combining current evidence with git history.
write_lessonWrite an evidence-backed lesson to the memory ledger. Loose remember-this notes are rejected.
BEFORE YOU INSTALL
7 tools when we connected.
npx on PATH
Both documented clients run the server on your machine.
CONTEXT COST
Rated by the size of the 7 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
npx @inferensys/contextful init --workspace .npx -y @inferensys/contextful servercodex mcp add contextful -- npx -y @inferensys/contextful server{
"mcpServers": {
"contextful": {
"command": "npx",
"args": [
"-y",
"@inferensys/contextful",
"server"
]
}
}
}