MemoryLens MCP: Letting AI Read a Memory Profile
Memory profiling in .NET has a discoverability problem. The tooling is excellent — dotMemory will tell you exactly which objects are surviving and who is holding them — but reading a retention graph is a skill, and the gap between “this List<T> is retained by an event handler” and “here is the line to change” is where most people give up.
I wanted my AI assistant to close that gap. So I built MemoryLens MCP.
What It Does
MemoryLens is an MCP server that wraps JetBrains dotMemory and exposes profiling as tools an agent can call: attach to a process, take a snapshot, analyse it, and compare two snapshots to see what grew between them.
The important part is what happens after the snapshot. Raw dotMemory output is a graph of object retention. MemoryLens runs it through a heuristic rule engine that maps common retention shapes onto their usual causes — undisposed subscriptions, static caches without eviction, closures capturing more than intended — and reports them as concrete suggestions rather than as a data structure to interpret.
Why an MCP Server
The point is not to make profiling faster. It is to make it something an agent will reach for unprompted.
When “take a snapshot and tell me what leaked” is a tool call rather than a separate application, a debugging conversation can include real measurements instead of speculation about what the allocation profile probably looks like. That changes the quality of the answer considerably.
Getting Started
Install as a .NET global tool:
dotnet tool install -g MemoryLens.McpIt is also published on npm and listed on Glama.ai. A local dotMemory installation is required — MemoryLens drives it, it does not replace it.
Source and tool reference: github.com/MarcelRoozekrans/memorylens-mcp