Context
context_pack
Returns a ranked, cited, token-budgeted bundle for the current task.
QueryBudgetScopeCitations
Case study / Context engineering
Context, Search Engine & Memory Layer
Contextful gives AI coding agents a local way to find the right files, compress the evidence, and remember decisions across sessions.
It runs as a CLI and MCP server for Codex, Claude Code, Cursor, Windsurf, GitHub Copilot, VS Code, Cline, Roo Code, Continue, and Zed.
TypeScriptNode.jsModel Context ProtocolSQLiteFTS5better-sqlite3Tree-sitter
Project
Contextful
Category
Context engineering, MCP server, local code search, and agent memory
Core loop
Index workspace, search code/docs/symbols, return context pack, store evidence-backed lessons
Primary surfaces
cxf CLI, MCP server, generated agent instructions, local state directory
Storage
SQLite, FTS5 search tables, graph tables, adjacency cache, memory ledger
Repository
github.com/Inferensys/contextful
Product problem
Agents need scoped evidence, not another broad file read.
Context packs
The agent asks one question and receives ranked citations inside a token budget.
Codebase search
Contextful searches code, docs, symbols, graph relationships, and saved memory through one local index.
Evidence-backed memory
A memory write needs file, symbol, or pack evidence so future sessions can trust the lesson.
Repository
Review the Contextful codebase.
The repo includes the TypeScript package, MCP server, CLI, SQLite schema, search engine, memory ledger, tests, docs, and npm package metadata.
github.com/Inferensys/contextfulSource repositoryAgent interface
The MCP tools are narrow on purpose.
Contextful gives agents the actions they need for project context without turning the MCP server into a general shell.
Search
search_code
Searches code, docs, symbols, and memory through the local index.
CodeDocsSymbolsMemory
Graph
trace_path
Traces file, symbol, module, and config relationships.
FilesSymbolsModulesConfig
Change risk
impact_analysis
Finds likely dependents and tests for a file, symbol, or module.
DependentsTestsRefsAffected area
Memory
write_lesson
Stores durable lessons only when the agent provides evidence references.
ClaimEvidence refsScopeStatus
Client support
Built for the coding tools teams already use.
The package is presented as a local MCP bridge for Codex, Claude Code, Cursor, Windsurf, GitHub Copilot, VS Code, Cline, Roo Code, Continue, and Zed.
The MCP server runs over stdio and can be started with npx.
Product surfaces
Context, search, and memory surfaces.
The diagrams show the problem Contextful addresses and the search model behind the product.
Context window
Keep the context window focused.
Contextful indexes the project locally and returns compact evidence instead of asking the agent to read broad file sets.
Context engineeringToken budgetCited evidence
Search engine
Search beyond plain grep.
The retrieval path combines lexical search, symbols, docs, graph relationships, and memory hits.
FTS5SymbolsGraph
System architecture
Local-first context infrastructure.
The system keeps source code on the developer machine while giving agents a structured retrieval and memory layer.
Index
Workspace indexer
Discovers files, skips ignored paths, reads content, and extracts chunks, symbols, and edges.
FilesChunksSymbolsEdges
Store
SQLite kernel
Stores files, chunks, graph nodes, graph edges, fingerprints, evidence packs, queries, and memories locally.
SQLiteFTS5Graph tablesWAL
Retrieval
Intent-aware search
Classifies queries for memory, impact, history, architecture, docs, exact lookup, or vague search.
MemoryImpactHistoryDocs
Output
Evidence packs
Saves context packs with citations, related files, symbols, graph paths, memory hits, confidence, and token estimates.
CitationsFilesSymbolsGraph paths
Memory
Memory ledger
Marks lessons stale when indexed files change, so old context does not silently carry forward.
Evidence refsStale stateSupersedesConfidence
Use cases
Where context engineering matters.
Contextful fits teams building coding agents, AI developer tools, internal copilots, and MCP-enabled engineering workflows.
Agents
AI coding agents
Give agents a repeatable way to fetch project context before editing code.
CodexClaude CodeCursorWindsurf
MCP
MCP developer tools
Expose code search, context packs, graph tracing, and memory through typed MCP tools.
MCP serverStdionpxTool schemas
Search
Enterprise code search
Support exact lookup and vague project questions with local codebase indexing.
Code searchDocs searchSymbol searchGraph search
Memory
Agent memory
Carry decisions, lessons, and project facts across coding sessions with evidence attached.
LessonsEvidence refsSession handoffStale checks
Retrieval
RAG for codebases
Use ranked local evidence as the retrieval layer for internal software engineering agents.
Context packsToken budgetCitationsLocal data
Operating loop
From workspace to cited context pack.
01
Initialize the workspace
The cxf CLI creates local Contextful state and generated agent instructions.
02
Index project context
The indexer records files, chunks, symbols, graph edges, language counts, git commit data, and warnings.
03
Ask for context
The coding agent calls context_pack or search_code with a query and optional token budget.
04
Return evidence
Contextful returns citations, file reasons, symbols, graph paths, memory hits, confidence, and token estimate.
05
Write memory with proof
Useful lessons can be saved only when the agent attaches valid evidence references.
FAQ
Contextful FAQ.
Answers for teams evaluating context engineering, MCP code search, agent memory, and local codebase indexing.
What is Contextful?
Contextful is a local context engine for AI coding agents. It indexes a workspace and exposes code search, context packs, graph tracing, impact analysis, git-history context, and evidence-backed memory through a CLI and MCP server.
How does Contextful help AI coding agents?
It gives agents a direct way to ask for ranked, cited project context instead of reading many files manually. That helps the agent start with better evidence before planning or editing code.
Is Contextful an MCP server?
Yes. Contextful ships an MCP server with tools for context_pack, search_code, trace_path, impact_analysis, why_changed, recall_memory, and write_lesson.
Does Contextful upload source code?
No. The current design is local-first. The README states that V1 does not call external embedding APIs, upload source code, edit source files, auto-fix code, or install dependencies inside the target workspace.
Is Contextful production software?
Contextful is early alpha open-source software. It is useful as a reference build for context engineering, MCP developer tools, local code search, and agent memory systems.
Output gallery
Project visuals.
Contextful cover, supported clients, context-window diagram, and retrieval model.
Contextful cover
Supported clients
Context window
Search model
Contact
Talk to the team about your AI system.
Share what you are building, where you need help, and what needs to ship next. We will reply with the right next step.
01
NDA available
We can start under NDA when the work requires it.
02
Direct team access
You speak directly with the team doing the technical work.
03
Clear next step
We reply with a practical recommendation on scope, implementation, or rollout.
30m
working session
Direct
team access