Skip to main content
Glama
akshaygp18

Memory Tracker MCP

by akshaygp18
OverviewSchemaRelated ServersScoreDiscussions
Python
Remote

Memory Tracker MCP

An MCP server that gives an AI assistant persistent memory, backed by an OpenAI vector store.

Memories are plain text. save_memory uploads each one as a file into a vector store named MEMORIES; search_memory runs a semantic search over that store and returns the matching chunks. The store is created on first use and reused after that, so memories persist across sessions and across clients.

Requirements

  • Python 3.14+

  • uv

  • An OpenAI API key

Related MCP server: MCP Memory

Setup

uv sync

Create a .env file in the project root:

OPENAI_API_KEY=sk-...

.env is gitignored. The server calls load_dotenv() at import, which resolves relative to the working directory — this is why the client configs below pass --directory.

Tools

Tool

Argument

Returns

save_memory

memory: str — the text to remember

{"status": "saved", "vector store id": ...}

search_memory

query: str — what to look for

{"results": [chunk, ...]}

Running it

Development, with the MCP Inspector:

uv run mcp dev server.py

Directly over stdio (what MCP clients do):

uv run python server.py

Client configuration

Claude Code

.mcp.json in this repo is picked up automatically when you start Claude Code in this directory. No further setup.

Claude Desktop

Add the block below to claude_desktop_config.json, then fully quit Claude Desktop (right-click the system tray icon → Quit — closing the window is not enough) and relaunch.

{
  "mcpServers": {
    "memory-tracker": {
      "command": "C:\\Users\\shivu\\.local\\bin\\uv.exe",
      "args": [
        "run",
        "--directory",
        "f:\\Agentic AI\\Memory_tracker_mcp",
        "python",
        "server.py"
      ]
    }
  }
}

Two things differ from the Claude Code config:

  • Absolute path to uv.exe. Claude Desktop launches servers with a minimal PATH that usually excludes ~\.local\bin, so a bare uv fails to spawn. Claude Code inherits your shell's PATH, so the short form works there.

  • Where the config file lives. For the standard installer it is %APPDATA%\Claude\claude_desktop_config.json. For the Microsoft Store (MSIX) build, AppData is redirected and the real path is:

    %LOCALAPPDATA%\Packages\Claude_pzs8sxrjxfjjc\LocalCache\Roaming\Claude\claude_desktop_config.json

    Editing the non-packaged path on a Store install has no effect. Reach it from the app instead via Settings → Developer → Edit Config.

Troubleshooting

Failed to build ... Expected a Python module at src\memory_tracker_mcp\__init__.py

pyproject.toml sets package = false under [tool.uv], which tells uv to treat this as a flat script project rather than build it as a package. Without it, every uv run tries to build an installable package and fails, because the server is a single server.py at the repo root and there is no src/ layout. Note that [project.scripts] still declares a memory_tracker_mcp:main entry point that does not exist — harmless while package = false is set, but it will break the build again if that line is ever removed.

Tools appear in the client but every call errors

Almost always a missing OPENAI_API_KEY. The --directory argument is what lets load_dotenv() find .env; drop it and the server still starts, but the OpenAI client has no key. As a fallback, pass the key through the config instead:

"env": { "OPENAI_API_KEY": "sk-..." }

That hardcodes the key into the config file, so prefer .env when it works.

Server shows as failed to start

Check the client's MCP log — for Claude Desktop, logs\mcp-server-memory-tracker.log in the same config directory. A spawn/ENOENT error means the uv.exe path is wrong; confirm it with where uv.

Notes

  • Every save_memory call writes a temp file with delete=False and opens it without closing the handle, so temp files accumulate in %TEMP%. Passing the text directly (file=("memory.txt", memory.encode())) would avoid the temp file entirely.

  • get_or_create_vector_store scans stores by name on every call, so each tool invocation costs an extra list request.

Available Tools

2 tools
save_memoryD
ParametersJSON Schema
NameRequiredDescriptionDefault
memoryYes

TDQS

D1/5.0
Behavior1/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Tool has no description.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness1/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Tool has no description.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness1/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Tool has no description.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters1/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Tool has no description.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose1/5

Does the description clearly state what the tool does and how it differs from similar tools?

Tool has no description.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines1/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Tool has no description.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

search_memoryC

Search memories in the vector store and return relevant chunks

ParametersJSON Schema
NameRequiredDescriptionDefault
queryYes

TDQS

C2.9/5.0
Behavior2/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

No annotations are provided, so the description must carry behavioral transparency. It only states that it searches and returns chunks, without disclosing details like result ordering, limits, or whether it has any side effects. This is minimal and relies on the agent's assumptions.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is a single sentence with no redundant content and the verb is front-loaded. It is concise and to the point, though it could have included more detail without becoming verbose.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness2/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

The tool is simple, but the description's 'return relevant chunks' is vague; it doesn't explain the format, size, or criteria of returned chunks. Since there is no output schema, the description should clarify the return value more fully.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters1/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The schema has one required 'query' string with no description, and the description provides no additional explanation of what the query should be or how it is interpreted. With 0% schema description coverage, the description fails to compensate, leaving the parameter's meaning entirely to its name.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description uses a specific verb 'search' and names the resource 'memories in the vector store', and clarifies the outcome 'return relevant chunks'. This clearly distinguishes it from the sibling 'save_memory', which is a write operation.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines2/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

There is no explicit guidance on when to use this tool versus save_memory. The description merely states what it does, leaving usage to implication rather than spelling out conditions or alternatives.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.

  1. 2 tool updatesv0.1.0
    • First observedsave_memory
    • First observedsearch_memory
Install Server
RayrunBuild, deploy, and operate MCP servers without becoming an infrastructure team.ray.runAd

TDQS

C2.3/5.0
Disambiguation5/5

The two tools, save_memory and search_memory, have clearly distinct purposes: one writes a memory and the other reads/retrieves them. No overlap or ambiguity exists.

Naming Consistency5/5

Both tool names follow a consistent verb_noun pattern (save_memory, search_memory), using the same resource noun. The naming is predictable and uniform.

Tool Count3/5

With only two tools, the set is at the low end of what is reasonable. For a memory tracker, save and search are core operations, so the count is borderline but acceptable.

Completeness3/5

The server covers the primary create and read/search operations, but it lacks update, delete, or list-all functionality. These are significant gaps for a complete memory management lifecycle.

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Connectors

Related MCP Servers

  • MCP Memory Tracker

    akaiserg
    F
    license
    B
    quality
    D
    maintenance
    A Model Context Protocol (MCP) server that provides persistent memory capabilities using OpenAI's vector stores, allowing AI assistants to save and search through memories across conversations.
    2
    -

  • Memory MCP

    gregpriday
    A
    license
    B
    quality
    D
    maintenance
    An MCP server that provides AI assistants with persistent, semantic memory using Turso for storage and OpenAI for vector search. It enables natural language operations to store, retrieve, and refine information with automatic duplicate detection and quality validation.
    5
    15
    MIT

  • Local Brain MCP

    ejsinfuego
    F
    license
    Not graded
    quality
    D
    maintenance
    A local MCP server for AI assistants to store and retrieve personal memories on disk, with optional semantic search using embeddings.
    -

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/akshaygp18/Memory-tracker-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server