Memtrace

Name: Memtrace
Author: Syncable

Local code graph with symbol history that coding agents query for search, impact and change tracking

Official · Syncable Rust Local · stdio Developer tools Knowledge and memory

GitHub npm Website

73 popularity score ★ 511 ↓ 1.6k/wk

Connect your client

Local (npm)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio

Opens the app, which asks before it adds the server.

Or add it by command or config

FirstTerminal

npm install -g memtrace
memtrace auth login

Claude CodeTerminal

claude mcp add memtrace -- memtrace mcp

Claude Desktopclaude_desktop_config.json

{
  "mcpServers": {
    "memtrace": {
      "command": "memtrace",
      "args": [
        "mcp"
      ]
    }
  }
}

Cursor~/.cursor/mcp.json

{
  "mcpServers": {
    "memtrace": {
      "command": "memtrace",
      "args": [
        "mcp"
      ]
    }
  }
}

VS Code.vscode/mcp.json

{
  "servers": {
    "memtrace": {
      "type": "stdio",
      "command": "memtrace",
      "args": [
        "mcp"
      ]
    }
  }
}

Codex~/.codex/config.toml

[mcp_servers.memtrace]
command = "memtrace"
args = ["mcp"]

Gemini CLITerminal

gemini mcp add --scope user memtrace memtrace -- mcp

Settings 4

`MEMTRACE_TELEMETRY`	Optional: Set to off (or 0/false/disabled/no) to stop the opt-out product telemetry.
`MEMTRACE_RAIL_SHADOW`	Optional: Set to off to stop only the search-quality measurement buckets in telemetry.
`MEMTRACE_DEFAULT_REPO`	Optional: Repo used by tools such as get_codebase_briefing when no repo_id is passed.
`MEMTRACE_OVERLAY_TTL_HOURS`	Optional: Age after which worktree overlays are cleaned up; 168 hours by default.

What it does

Memtrace is the MCP server of a closed-source code-intelligence product from Syncable. It parses a repository with Tree-sitter into a local graph of symbols (functions, classes, types, endpoints) and their links (calls, imports, implements), keeps the version history of each symbol from git, and serves that graph to a coding agent over stdio through `memtrace mcp`. The agent can then search code, follow callers and callees, estimate what a change would break, see what changed over a period, and map HTTP calls between services, instead of re-reading files each session.

Parsing and embeddings run on the user's machine; no LLM is called to build the graph. Beyond plain code queries it carries quality checks (dead code, duplicates, complexity, hotspots), a pull-request reviewer for GitHub, tools for several agents to announce edits and take leases on the same repo, and six "Cortex" tools that recall recorded design decisions and the code they produced.

The README advertises "25+" tools; the docs' tool reference says `memtrace mcp` registers 90: 81 core graph and runtime tools, three hosted-docs tools and six Cortex tools. 38 of the 81 core tools draw on the plan's query quota, and when the quota runs out a metered tool still returns a normal success result whose payload is a quota-error JSON object, not a protocol error (docs: MCP tool reference). The product is in private beta behind a waitlist and needs a signed-in Memtrace account; the licence is a proprietary EULA.

What it needs

Runtime	Node.js 18+

Node.js 18 or later and npm, installed globally with `npm install -g memtrace`, which pulls a prebuilt binary for macOS on Apple Silicon, Linux x64 or Windows x64. A Memtrace account: the product is in private beta (waitlist at memtrace.io) and refuses to run anonymously; sign in once with `memtrace auth login` or on the first `memtrace start`. Git for the history features. The README asks for at least 4 cores, 8 GB RAM and 5 GB free disk. The Cortex tools need the sidecar that `memtrace start` launches.

Limits

Closed source and in private beta; access is rolled out from a waitlist. Free for individual developers per its README, but 38 core tools are metered against a query quota (the teams plan is unmetered), and an exhausted quota shows up as an error payload inside a successful result. Built for code: it does not keep conversational memory. The first index is CPU and RAM heavy and the first run downloads an embedding model. The README still lists `execute_cypher`, which the current tool reference does not include. Nightly builds are marked as not always stable.

Who it suits

Suits developers who run one or several coding agents on a large or multi-service codebase and want callers, blast radius and change history answered from a local index rather than file reads, and who accept a closed-source binary, an account and phone-home licence checks. Pick an open-source code-graph or plain file-search server if you cannot sign up for the beta, need fully offline use, or want to audit the code. For memory of conversations or user facts rather than code, its own README points out it is not built for that.

Tools 55 of 90

index_directory: Parse a local folder into the code graph, resolve cross-file links and embed symbols; returns job ids
list_indexed_repositories: List indexed repos with node and edge counts, branch and last index time
delete_repository: Remove every node, edge, episode and embedding stored for one repo
check_job_status: Report which stage a background indexing job has reached
list_jobs: List background indexing jobs
get_repository_stats: Counts of nodes, edges and symbol kinds for an indexed repo
watch_directory: Keep a folder under watch and re-index changes as they happen
list_watched_paths: Show which folders are being watched
unwatch_directory: Stop watching a folder
list_worktrees: List git worktree overlays held for a repo
cleanup_worktrees: Drop stale worktree overlays (missing, merged, or older than the TTL)
find_code: Hybrid keyword and semantic code search, optionally across all repos or at a past point in time
find_symbol: Look up a symbol by exact or fuzzy name
get_source_window: Return a slice of a source file around given lines, with optional compression
get_directory_tree: Map a directory tree in one call, with depth and entry caps
analyze_relationships: Callers, callees, class hierarchy or type usages of a symbol
get_symbol_context: One-call overview of a symbol: definition, neighbours and risk
get_impact: Estimate the blast radius of changing a symbol, with a risk level
find_dead_code: Find symbols nothing calls
find_duplicate_code: Find duplicated code blocks
calculate_cyclomatic_complexity: Compute cyclomatic complexity for code
find_most_complex_functions: Rank functions by complexity
get_function_quality_metrics: Quality figures for one function
find_hotspots: Find code that changes often within a time window
find_code_review_issues: Run graph-based review checks over changed code
review_github_pr: Review a GitHub pull request against the local index and optionally post findings to GitHub
get_daily_briefing: Summary of what changed between two anchors
create_anchor: Mark a named point in a repo's history for later comparison
get_timeline: Version history of a symbol
get_evolution: Rank what changed over a period using one of six scoring modes
detect_changes: Map a diff or changed files to the symbols they touch
get_changes_since: List changes since a given episode or time
get_cochange_context: Show which code tends to change together with a symbol
get_episode_replay: Replay a recorded change episode
record_external_episode: Add an agent- or externally-authored episode to a repo's timeline
find_api_endpoints: List HTTP endpoints found in indexed code
find_api_calls: List outgoing HTTP calls found in indexed code
get_api_topology: Graph of which services call which endpoints across repos
link_repositories: Add a typed link between two repos the HTTP linker cannot infer
ingest_graph_fragment: Add nodes and edges from generated code or other sources the parsers miss
list_processes: List traced execution flows
get_process_flow: Show the steps of one traced flow
list_communities: List module clusters found by community detection
find_central_symbols: Rank symbols by centrality
find_bridge_symbols: Find symbols that connect otherwise separate parts of the code
find_dependency_path: Find the dependency path between two symbols
get_codebase_briefing: Compact start-of-session overview of a repo's structure and risks
fleet_publish_intent: Announce an intended edit to other agents and see overlapping work
fleet_status: Show live intents and conflicts among coordinating agents
fleet_acquire_lease: Take a time-limited lease on code before editing it
fleet_release_lease: Release an edit lease
recall_decision: Look up recorded design decisions matching a free-text query
verify_intent: Check whether a recorded decision held across the changes that implemented it
why_is_this_here: Trace the decision lineage behind a symbol
governing_rules: List governance rules in force for a file path

Popularity

73 popularity score, on our list (the bar is 36)

★ 511 ↓ 1.6k/wk

GitHub stars	511	32.5 points
Stars in 30 days	+22	6.8 points
npm downloads, last week	1,563	25.6 points
Official server	Yes	8 points

Measured 2026-11-07. How the score works.

Where it is published

Code	syncable-dev/memtrace-public ★ 511
npm	memtrace 1.2.17, 2026-11-04

What it can reach

Runs	On your machine
Reaches	Files on your machine, Your code, The internet, Your account at a service
Changes things	Yes: it can write, send, create or delete
Sign-in	Account credentials
Licence	MEMTRACE END USER LICENSE AGREEMENT (EULA) (its licence file)
Last code change	2026-11-04 (a release)

The binary runs on your machine and reads the folders you index, storing the graph, embeddings and history in a local database it can also delete from. Per PRIVACY.md it sends the licence key and machine hostname to memtrace.io on start, a usage heartbeat every 15 minutes (aggregate counts, no names or source), and opt-out telemetry every 60 seconds; it downloads embedding weights from Hugging Face once. review_github_pr fetches a PR diff with a short-lived GitHub App token minted by memtrace.io and can post comments to GitHub. `npm install -g` also writes MCP config and skills into the config folders of several agents. The indexer is closed source.

Over time

GitHub stars 511 +22 (+4 %)

511489

9 Oct 20267 Nov 2026

npm downloads a week 1.6k −55 (−3 %)

7.5k143

20 Apr 20267 Nov 2026

Before the site measured it, read from npm's daily download counts.

Popularity score 73 +16 (+29 %)

7356

9 Oct 20267 Nov 2026

What is rising across the directory

Sources

Page checked 2026-10-14.

Badge for your README

Shows this server's rank in its category, and stars for a server that runs on your machine. It updates every hour.

MarkdownREADME.md

[![Memtrace on EveryMCPServer](https://everymcpserver.com/badge/memtrace-public.svg)](https://everymcpserver.com/servers/memtrace-public)

Connect your client

Add in one click

Or add it by command or config

Settings 4

What it does

What it needs

Limits

Who it suits

Tools 55 of 90

Similar servers

trace-mcp

OpenLore

projectmem

code-review-graph

GitNexus

Engram