Articles & Guides
The complete index of trace-mcp documentation, architectural deep dives, benchmarks, and head-to-head comparisons against other tools in the code-intelligence ecosystem.
VS Comparisons
MCP Code-Graph Tools Compared: Serena, Repomix + 20 More
Compare 20+ MCP code-graph tools head-to-head: token costs, 81 languages, 88 frameworks, GitHub stars. Pick the one that fits your stack.
Repomix Alternative: trace-mcp vs Repomix for AI code context
Repomix packs your repository into one prompt; trace-mcp indexes it into a queryable graph. Head-to-head on token cost, freshness, search, refactoring.
Serena MCP Alternative: trace-mcp vs Serena for agent code navigation
Serena drives a live language server; trace-mcp precomputes a framework-aware code graph. Head-to-head on precision, startup cost, refactoring, memory.
codebase-memory-mcp Alternative: trace-mcp vs codebase-memory-mcp
Both build a persistent code knowledge graph for AI agents. Head-to-head on language coverage, tool cost, framework awareness, refactoring and security.
CodeGraphContext Alternative: trace-mcp vs CodeGraphContext for AI agents
CodeGraphContext drives 11 SCIP indexers into a graph database you choose. trace-mcp ships one embedded store, framework edges and a write path.
SocratiCode Alternative: trace-mcp vs SocratiCode for AI agents
SocratiCode pairs Qdrant vector search with ast-grep in Docker. trace-mcp runs locally with SQLite, 81 languages, framework edges, and refactoring.
jCodeMunch Alternative: trace-mcp vs jCodeMunch for AI agents
Looking for a jCodeMunch alternative you can use commercially? Honest trace-mcp vs jCodeMunch comparison: license, framework edges, refactoring.
TokenSave Alternative: trace-mcp vs TokenSave for AI agents
TokenSave offers 86 MCP tools in Rust under MIT. trace-mcp adds 88 framework integrations, AST refactoring, and OWASP taint analysis.
GitNexus Alternative: trace-mcp vs GitNexus for AI agents
GitNexus couples LadybugDB with in-memory Leiden clustering under a non-commercial license. trace-mcp adds 88 frameworks, typed tools, and AST refactoring.
Graft Alternative: trace-mcp vs Graft for AI agents
Graft builds a markdown cache with paid LLM APIs. trace-mcp indexes code locally with zero token bills, typed tools, 88 frameworks, and AST refactoring.
Narsil-MCP Alternative: trace-mcp vs Narsil-MCP for AI agents
Narsil-MCP offers 90 MCP tools and SPARQL RDF in Rust. trace-mcp adds 88 framework integrations, AST refactoring write tools, and OWASP taint analysis.
CodeGraph MCP Alternative: trace-mcp vs codegraph for AI coding agents
codegraph ships one orientation tool; trace-mcp ships a broad graph with refactoring, security and memory. Head-to-head on tools, coverage, benchmarks.
Context Mode alternative? trace-mcp vs Context Mode for AI coding agents
Context Mode keeps raw tool output out of the context window; trace-mcp makes questions about your code cheap to ask. Head-to-head — and why run both.
code-review-graph Alternative: trace-mcp vs code-review-graph
code-review-graph builds an incremental SQLite graph; trace-mcp adds framework awareness, refactoring, security and memory. Head-to-head comparison.
Repomix vs codegraph: packing a repo vs indexing it for AI agents
Repomix packs a repo into one file an agent reads; codegraph indexes it into a graph an agent queries. Head-to-head on cost, freshness and benchmarks.
ast-grep Alternative: trace-mcp vs ast-grep for AI code context
ast-grep finds code by AST pattern; trace-mcp indexes it into a queryable graph. Search, rewriting, impact analysis, MCP use compared.
Continue Alternative: trace-mcp vs Continue for AI coding agents
Continue is an open-source coding agent for VS Code, JetBrains and terminal. trace-mcp is the persistent code graph any agent queries. Compared.
Aider Alternative: trace-mcp vs Aider for AI pair programming
Aider pair-programs in your terminal with a repo map; trace-mcp serves a persistent code graph over MCP. Map vs graph, git loop, memory compared.
Cody Alternative: trace-mcp vs Sourcegraph Cody for code context
Cody is enterprise-only cloud search plus assistant. trace-mcp is the local-first graph alternative — no key, no cloud, framework edges compared.
ripwire Alternative: trace-mcp vs ripwire for agent code maps
ripwire is the zero-dependency CLI map for agents with published evals; trace-mcp is the persistent MCP graph. Map vs graph, honesty, memory compared.
IDE Context Alternative: trace-mcp vs Cursor rules, Copilot setup
Cursor rules and Copilot instructions steer one IDE session; trace-mcp is the persistent graph every agent queries. Session context vs stored edges.
Benchmarks
PR Review Context Benchmark — Measured Input-Token Cost on Real Pull Requests
Reproducible measurement of the input tokens trace-mcp context saves over naive file loading when reviewing real merged pull requests in open-source repos.
TOON Output Format — Measured Token Savings on Real Tool Calls
Real-world token measurements for trace-mcp's TOON (Token-Oriented Object Notation) output format across tabular MCP tool responses.
Tool response token cost
What trace-mcp tool responses cost in tokens, per tool, weighted by real call volume — including the ones that cost more than the reads they replace.
Guides
What trace init installs — guard hooks, Read/Bash mirrors and client config
What trace init writes to your machine and how to audit it: guard hooks, opt-in Read/Bash mirrors, the routing block, client config, tweakcc tiers.
How to Reduce Claude Code Token Usage — 7 measured tactics
Seven ways to cut token usage in Claude Code, ordered by measured impact: stop full-file reads, trim your MCP tool surface, pick the output format.
trace-mcp Configuration Guide — .trace.json Reference
Configure trace-mcp in .trace.json: indexing, presets for 10 workflows, LSP servers, quality gates. Copy a working example — every key stays optional.
System Prompt Routing via tweakcc — pairing trace-mcp with Claude Code
How to pair trace-mcp with tweakcc to patch Claude Code's system prompts and route tool selection toward trace-mcp instead of native file reads.
Contributing to trace-mcp — local setup, build, and test
How to set up trace-mcp for local development: install, build, run the test suite, and the conventions to follow before opening a PR.
Frameworks
Supported Languages & Frameworks — 81 languages, 88 framework integrations
Full list of languages and frameworks trace-mcp understands out of the box — web frameworks, ORMs, UI libraries, and tooling, across 81 languages.
Language capability matrix: what 81 languages extract
What each of the 81 languages trace-mcp supports actually extracts — parser, edge kinds resolved and test coverage, one row per language.
Reference
MCP Tools Reference — code-intelligence tools by task and framework
The trace-mcp MCP tools you reach for most, grouped by task: navigation, refactoring, impact analysis, security and framework-aware queries.
Tool index — every MCP tool the server registers
Alphabetical index of every trace-mcp MCP tool, generated from the server's own registrations: what each one does and when it is available.
Config index — every key trace-mcp accepts
Complete index of every trace-mcp configuration key, generated from the schema itself: type, allowed values and the real default.
Quality Gates — configure trace-mcp's thresholds
The eight quality_gates.rules keys trace-mcp checks, which three run by default, and how this project calibrated its own thresholds as a worked example.
Concepts
Code Graph MCP Server for AI Coding Agents
How a code graph MCP server gives AI coding agents persistent symbol, call, and framework relationships without re-reading whole repositories.
Architecture — indexing pipeline, storage, and MCP server internals
How trace-mcp indexes a codebase into a queryable graph: tree-sitter parsing, SQLite + FTS5 storage, optional LSP enrichment, and the MCP server on top.
Session Analytics & Coverage Intelligence — token savings, wasteful patterns
trace-mcp's built-in analytics engine parses AI agent session logs, tracks token savings, detects wasteful patterns, and assesses technology coverage.
Decision Memory — a persistent knowledge graph for architectural decisions
trace-mcp's decision graph captures architectural decisions, tech choices, bug root causes and conventions — each linked to the code it is about.
Daemon memory: what it costs and what caps it
Measured resident-set attribution for the trace-mcp daemon — what each region holds and which config knob bounds it.
MCP Tracing — OpenTelemetry spans for every MCP tool call
How to trace MCP tool calls: trace-mcp emits OpenTelemetry spans for every MCP tool invocation and AI provider call, exported to Jaeger or Langfuse.
Privacy — what trace-mcp sends, and how to turn it off
Everything trace-mcp sends off your machine: one anonymous daily ping, its complete field list, both opt-outs, and how to delete the local state it keeps.
Last updated: October 19, 2026