Let Claude Code read your notes with MCP

Mac utility developer Mohak Bajaj released a local Model Context Protocol integration for Jot, enabling command-line AI coding assistants to interact directly with local developer notes. Detailed in sayjot.xyz's report, the update introduces an embedded server executable that bridges Claude Code, Codex, and Cursor directly to markdown capture files on macOS without requiring online account registration or external web service tokens.

This report examines how software engineers can register the local toolset across their development environments, how atomic filesystem operations safeguard note consistency, and why the system intentionally restricts agent access to destructive file removal commands. It is written for developers looking to automate personal knowledge retrieval during terminal coding sessions without risking data loss or compromising local security boundaries.

01Local storage architecture and markdown files

Jot organizes captured developer context by storing every entry as a distinct Markdown document inside a designated filesystem directory, defaulting to the path ~/Documents/Jot. Users create entries across the operating system by highlighting text in any active window and pressing the Shift key twice, or by executing commands directly in a terminal shell. The background capture application does not need to remain active for external processes to access the directory.

The companion jot mcp utility operates as a self-contained local server communicating strictly over standard input and output streams on the host machine. Because the architecture relies on direct filesystem access rather than cloud APIs, personal notes remain entirely on local hardware until an AI assistant explicitly includes their text inside an outbound prompt context. The execution model requires no centralized accounts, remote servers, or credential management.

02Tool catalog and atomic write operations

The server exposes seven functional tools designed to give AI agents structured access to developer notes without corrupting stored data. Read-only commands include list_jots for filtering records by tags, sections, or completion states, search_jots for case-insensitive keyword searches across open entries, and read_jot for extracting complete front matter and image paths. Write operations permit agents to generate new jots, toggle completion statuses, update metadata tags, and move processed items into archive storage.

All agent-driven modifications utilize atomic write procedures directly on the underlying Markdown files within the vault. The desktop application continuously observes directory events, immediately reflecting external additions in the user inbox without manual refresh operations. Environments supporting experimental MCP Apps gain access to an inbox viewer alongside an eighth tool named jot_image that resolves embedded graphics.

03Archive safeguards against accidental note deletion

The tool architecture intentionally omits any facility for deleting stored notes through automated agent commands. Autonomous agents act quickly and occasionally misunderstand instructions, making broad cleanup prompts a substantial liability for critical records such as debugging traces or API references. By replacing permanent deletion capabilities with archive status changes, the server ensures that misclassified notes merely leave the immediate inbox while preserving the underlying files.

If an AI assistant mistakenly flags an essential entry as finished or archived, human operators can quickly restore the record from the local directory or application interface. File deletion remains an exclusively manual human operation managed through the graphical application, which routes removed records into a dedicated system trash folder prior to permanent purge. This structural boundary protects developers from irreversible context loss during unattended agent refactoring sessions.

04Terminal commands and multi client configuration

Installing the server begins by downloading the Jot application archive from sayjot.xyz/download and launching the package once on macOS. The executable binary resides within the application bundle at /Applications/Jot.app/Contents/Helpers/jot, eliminating external package dependencies. Developers can create a system-wide symlink through the in-app settings under Vault & Agents or by running the standard linking command in their terminal.

ln -sf /Applications/Jot.app/Contents/Helpers/jot /usr/local/bin/jot

To link the toolset across all Claude Code workspaces, developers run the user-scoped registration command and verify tool availability using the built-in inspection command. Claude Code updates can be fetched using claude update, while alternative programming environments such as Codex and Cursor support equivalent configuration routines.

claude update
claude mcp add --scope user jot -- /usr/local/bin/jot mcp

codex mcp add jot -- /usr/local/bin/jot mcp

{
 "mcpServers": {
 "jot": { "command": "/usr/local/bin/jot", "args": ["mcp"] }
 }
}

/mcp

05Custom vault paths and plain language workflows

Developers interact with their notes through natural conversational queries rather than by invoking specific tool signatures directly. Inquiries instructing Claude Code to inspect project tasks, summarize documentation excerpts, or append follow-up items from code reviews automatically trigger the appropriate underlying routines. Short and precise search terms provide the most reliable substring matches across open records, while historical queries require explicitly targeting archived files.

The server defaults to reading the path specified in application settings but supports dynamic redirection using the --vault command flag or the JOT_VAULT environment variable. This redirection allows developers to target directories located inside existing Obsidian vaults, granting agents direct visibility into established cross-linked documentation. For file-only agents, Jot maintains INDEX.md and SKILL.md reference files in the root folder alongside an optional skill installer.

Jot requires macOS 12 or later and operates natively across Apple silicon and Intel processors. Complete tool definitions are published in the open-source GitHub repository at MohakBajaj/jot-mcp for community auditing. The software provides a 3-day evaluation window before requiring a $29 single-user license, with an early promotion discounting the first 50 licenses to $9 including bundled server capabilities.