Memtrace MCP
A local code graph with version history: an MCP server that tells the agent who calls a function and what an edit breaks
- 508 ★ on the repository
- Last commit
Extended access
Why this level
- Indexes the whole chosen code directory and stores the graph on disk
- A closed binary from npm that validates its license over the network
- The installer writes skills and MCP configuration into several agents at once
Install
- Works in
Manual install
npm install -g memtraceOne command installs the binary, 17 skills, the MCP server, the plugin and the marketplace for Claude Code and Claude Desktop. Private beta access is required.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Indexes the whole chosen code directory and stores the graph on disk
- A closed binary from npm that validates its license over the network
- The installer writes skills and MCP configuration into several agents at once
This is third-party code. Review the repository files before installing.
What it does
Memtrace indexes a codebase into a graph: functions, classes, interfaces, types and endpoints become nodes, while calls, imports, implementations and containment become edges. Parsing runs locally through Tree-sitter on Rust, with no language model calls, so indexing costs nothing in API fees. Every symbol keeps its version history, and the agent asks questions over time: what changed this week, what broke the usual pattern, what will an edit hurt. The server exposes more than 25 MCP tools. Search works by meaning and by text (find_code, find_symbol), relationships show callers and callees (analyze_relationships), and get_impact computes the blast radius of a change with a severity rating. There is also dead code detection, complexity hotspots, bridge and central symbols, module detection with the Louvain algorithm, and free-form Cypher queries. A separate layer maps HTTP calls between repositories: which services hit which endpoints. Frameworks such as FastAPI, Express, NestJS, Django, Gin and Rails have dedicated scanners, and infrastructure is covered by parsers for Terraform, GitHub Actions, Helm and Kubernetes manifests. The server ships with 17 skills. Nine teach the agent to use one tool each, and eight chain tools into workflows: codebase exploration, change impact analysis, incident investigation, refactoring, episode replay.
Who it is for. For teams and solo developers who change large or multi-service repositories with Claude Code, Cursor or Codex and want to see the consequences of an edit up front.
Good fit when
- Before a refactor you need to know which functions and tests depend on the code being changed
- Incident investigation: you need the symbols that changed near the moment of failure
- Several agents edit one repository and should share one call graph and history
- In a microservice setup you need to find which services call a specific endpoint
- You want dead code and the most complex functions listed before a tech-debt cleanup
Not a fit when
- You need memory of conversations, preferences and facts about people: Memtrace stores code structure and does not parse dialogue
- You require an open license and open indexer source: the indexer and the MemDB database are closed and the license is proprietary
- The machine has fewer than 4 cores and 8 GB of memory: the first index is heavy on CPU and RAM
- You need access right away: the service is in a private beta with a waitlist
Example request
What breaks if I change the signature of the payment creation function? Show the callers and the tests the edit affectsMCP
- Transport
- stdio
- Authentication
- not required
Environment variables
MEMTRACE_TELEMETRY- The value off disables crash reports and usage counters.
MEMTRACE_BIN- Path to the binary when you want to pin a local install, for example for DeepSeek Harness.
Limitations
The license is proprietary (EULA). The tool is free for individual developers during the private beta and after general availability; the indexer and the MemDB database are closed, and only the benchmark suite is open under MIT. Access is granted in batches, so you have to join the waitlist on memtrace.io.
Minimum to run: 4 cores, 8 GB of memory, 5 GB of disk and Node.js 18. Temporal analysis needs git with full history. The npm package installs a small main module plus a platform binary (macOS on Apple Silicon, Linux x64 or Windows x64). If npm skips optional dependencies, run the install with the --include=optional flag.
Source code never leaves the machine. The only network traffic is license validation, aggregate node and edge counts and crash reports, which MEMTRACE_TELEMETRY=off disables. Speed and accuracy figures in the docs were measured by the author on their own suites, so check them on your own code.
How to disable. Run `memtrace uninstall`: it removes the skills, the MCP server, the plugin and the settings. Then remove the package with `npm uninstall -g memtrace`. If the package is already gone, run `node ~/.memtrace/uninstall.js`.
FAQ
How do I install Memtrace in Claude Code?
`npm install -g memtrace` installs the binary, 17 skills, the MCP server, the plugin and the marketplace for Claude Code and Claude Desktop. The manual equivalent is `claude plugin marketplace add https://github.com/syncable-dev/memtrace-public.git`, `claude plugin install memtrace-skills@memtrace --scope user` and `claude mcp add memtrace -- memtrace mcp`.
Does Memtrace send my code anywhere?
No, indexing and queries run on your machine. The only network traffic is license validation, aggregate node and edge counts and crash reports without source code, file paths or symbol names. `MEMTRACE_TELEMETRY=off` turns telemetry off.
How is Memtrace different from Mem0 and Graphiti?
Mem0 and Graphiti build a graph from dialogue and extract entities with a language model. Memtrace parses code with Tree-sitter and makes no model calls, so code indexing carries no API cost, but it does not keep conversational memory.
Which agents does Memtrace work with?
The installer configures Claude Code, Claude Desktop, Cursor 2.4 and newer, Codex CLI, Windsurf, VS Code with Copilot, Hermes, OpenCode and Kiro. For Cline, Roo Code and any other MCP client you add the server by hand with the `memtrace mcp` command.
Is Memtrace free?
It is free for individual developers during the private beta and after general availability. The license is proprietary: the indexer and the database are closed. To get access, join the waitlist on memtrace.io.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server that builds a code graph of functions, classes, calls and routes and answers the agent's structural queries locally
An MCP server that builds a code graph once and serves the answer to the agent instead of re-reading files
A dependency-free CLI and MCP server that builds a call graph of a repository so the agent understands code without reading whole files
Local code structure graph: builds a repository map with tree-sitter and serves the assistant only the context a change actually touches over MCP
An MCP server that turns a large codebase into a semantic feature graph for precise navigation and meaning-based search
An MCP server and CLI for code research: search and understand your project and all of GitHub with fewer tokens
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Memtrace (this entry) | yes | not required | 508 | |
| codebase-memory-mcp | no | not required | 47,740 | a local code graph for structural queries, without version history or temporal scoring |
| CodeGraph | no | not required | 75,798 | entry points and blast radius in one call, without change chronology or a cross-service call map |
| Mem0 | no | 68,059 | remembers conversations and facts and does not parse code structure |
README in short
The documentation explains the bi-temporal graph idea and describes six change-scoring modes: compound, impact, novelty, recency, directional and overview. Separate sections list languages, framework scanners, an agent compatibility table and hardware requirements.