Atlas / MCP servers / egroup-labs / Kept

KeptCAUTION

Name: Kept
Author: egroup-labs

mcp/egroup-labs/kept

MCP server by egroup-labs · chrome extension · claude code · knowledge graph · Audited 2026-10-07

⚑ Report / re-scan Source ↗

Search, archive, and recall your AI conversations. ChatGPT, Claude, Gemini, Grok, Kimi. Local-first, MIT licensed.

Verdict

CAUTION

Grade

Trust score

81 /100

Exposed tools

34 27r · 5w · 2d

Transport

stdio

License

MIT

Stars

144

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

Kept saves your AI conversations as local Markdown files, then gives you a desktop app to search, browse, connect, and reuse them.

It works with ChatGPT, Claude, Gemini, Grok, and Kimi. Your archive lives on your machine under ~/.kept/, with an Obsidian-compatible vault plus local indexes for full-text search, topics, projects, and graph views.

Quick Install | Download | Setup | Build from source | MCP server

Why Kept

AI chats often become working memory: debugging trails, research notes, product decisions, prompts, snippets, and half-finished ideas. Most of that history stays inside vendor UIs.

Kept turns it into files you own.

Plain Markdown, grouped by provider.
Fast local search with SQLite FTS5.
Graph and topic views for finding connections across old conversations.
Optional chat over your own archive using the model provider you configure.
An MCP server so coding agents can read, search, and manage the vault.

How It Works

Chromium browser extension
-> reads conversations from provider API endpoints using your signed-in session
-> normalizes messages and supported image assets
-> sends them to the Kept desktop app on http://localhost:18241

Kept desktop app
-> writes Markdown files to ~/.k

Read from source at commit 589f2199a7ffOBSERVED · 2026-10-07

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control.

claude-code

claude mcp add kept-vault-server -- npx -y [email protected]

claude-desktop

{
  "mcpServers": {
    "kept-vault-server": {
      "command": "npx",
      "args": [
        "-y",
        "[email protected]"
      ]
    }
  }
}

Exposed tools (34)

27 read · 5 write · 2 destructive. Blast radius: 2 tools can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.

Tool	Risk	Description
`Anthropic`	read	Claude models for chat and knowledge graph extraction
`OpenAI`	read	GPT models for chat and knowledge graph extraction
`OpenRouter`	read	Route to 300+ models from any provider
`commit`	write	Create a git commit with conventional format
`delete_file`	destructive	Delete a file from the vault. Cannot delete directories.
`deploy`	write	Deploy to production
`execute_code`	write	Run a python, javascript, or shell snippet on the user
`get_neighbors`	read	Get all directly connected nodes and edges for a given node ID.
`get_stats`	read	Get aggregate statistics about the knowledge graph: entity count, triple count, conversation count, project count, and top entities.
`grep_knowledge_files`	read	Search knowledge base files using a regex pattern. Optionally scope to a single file. Returns matching lines with file path and line number.
`grep_vault`	read	Regex search across all markdown files in the vault
`highlight_nodes`	read	Highlight specific nodes in the knowledge graph view by emitting a frontend event.
`list_conversations`	read	List all archived conversations with metadata (title, platform, model, date, message count). Optionally filter by platform.
`list_directory`	read	List contents of a directory.
`list_fs_allowed_paths`	read	List the filesystem paths the agent is allowed to access.
`list_knowledge_files`	read	List all files and directories in the user
`list_nodes`	read	List nodes in the knowledge graph, optionally filtered by node type.
`list_vault`	read	List all files and directories in the vault root
`move_file`	write	Move or rename a file within the vault
`read_conversation`	read	Read the full markdown content of a specific conversation by its file path. Use this after searching to read conversations in detail.
`read_file`	read	Read a file from the local filesystem (scoped to allowed directories).
`read_image`	read	Describe or extract text from an image file.
`read_knowledge_file`	read	Read the full text content of a file from the knowledge base. Use list_knowledge_files first to discover available files.
`read_pdf`	read	Extract text content from a PDF file (scoped to allowed directories).
`read_web_page`	read	Fetch and extract readable text from a public HTTP(S) web page. Use after web_search when a result needs verification or more detail.
`recommend_conversation`	read	Recommend a conversation to link to the project. Call this for each relevant conversation you find.
`review`	read	Review code changes
`search_conversation_content`	read	Full-text search across archived conversations. Optionally filter by platform or title. Returns deduplicated results with cleaned snippets.
`search_conversations`	read	Full-text search across all archived conversations. Returns matching conversation titles, snippets, and file paths. Use FTS5 query syntax: simple words are OR
`search_knowledge_files`	read	Case-insensitive text search across all knowledge base files. Returns matching lines with file path, line number, and content.
`search_nodes`	read	Search for nodes in the knowledge graph by keyword. Returns matching nodes and their edges.
`update_file`	destructive	Overwrite an existing file in the vault. Fails if file does not exist.
`web_search`	read	Search the public web for current or source-backed information. Returns title, URL, and snippet results. Use this for current events, recent facts, product/library/version lookups, or when the user asks to search the web.
`write_file`	write	Create a new file in the vault. Fails if file already exists.

Trust audit

CAUTIONgrade B · trust 81/100 Install with care. The audit found things worth knowing before you trust its output.

Layer	What it checks	Result
L0	Provenance & inventory	WARN
L1	Static analysis of the code	WARN
L2	Instruction surface (what it tells the agent)	PASS
L3	Class-specific surface	WARN
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (5 observation(s))
Network: declared (15 observation(s))
Shell: none-observed
Dependencies: not all pinned
Secrets in source: none-found

Findings (19)

MEDIUMInventory / provenance · inv.binary · CWE-1104

app/src-tauri/icons/icon.icns

icon.icns

Why it matters. a compiled or binary member cannot be reviewed from source

Fix. ship source, or explain the binary in the README

MEDIUMInformation disclosure · disclose.log_secret · CWE-209, CWE-532

extension/connect.js:27

console.log("[Kept connect.js] token from meta tag:", token ? token.slice(0, 8) + "..." : "MISSING");