Provides access to Cursor IDE's conversation history stored in SQLite databases, enabling querying, analysis, and export of chat data across workspaces
cursor-db-mcp (MCP Server)
An MCP server for querying, exporting, and analyzing Cursor IDE conversation history from the new workspaceStorage
(SQLite) layout.
Highlights
- Query conversations across all workspaces (SQLite
state.vscdb
underworkspaceStorage
) - Search and analyze messages, code blocks, and estimated code changes
- Export data in
json
,csv
, ormarkdown
- Workspace utilities to list/filter recent workspaces
- Diagnostics tool to discover actual storage keys in your environment
Data sources and fallbacks
- Primary: legacy chat key (if present):
workbench.panel.aichat.view.aichat.chatdata
- Heuristics: if the legacy key is missing, the server scans values for structures containing
conversations/messages/assistant/role
. - Last resort: synthesizes conversations from
aiService.generations
(user-side prompts). Optionally, it heuristically pairs items fromaiService.prompts
as assistant replies to improve completeness.
Note: Depending on your Cursor version, assistant replies may not be locally persisted under known keys. The diagnostics tool can help you discover new keys. If you can share a confirmed key/path containing assistant messages, this server can be adapted quickly.
Installation
Run MCP Inspector for local testing:
The Inspector prints a local URL for interactive testing in your browser.
Configure in Cursor
Basic configuration:
Custom workspaceStorage
path:
Available Tools
- list_workspaces
- Input:
{ recent_days?: number }
- Returns workspace list with hash, path, projectPath, lastModified
- Input:
- get_workspace_conversations
- Input:
{ workspace_hash: string }
- Use
list_workspaces
to obtainworkspace_hash
- Returns conversations for that workspace
- Input:
- get_all_conversations
- Input:
{ limit?: number }
- Returns cross-workspace conversations (sorted by updated time)
- Input:
- search_conversations
- Input:
{ query: string, limit?: number }
- Full-text search in titles and messages
- Input:
- analyze_conversation
- Input:
{ conversation_id: string }
- Returns stats (message counts, code blocks, estimated code changes)
- Input:
- export_conversations
- Input:
{ format?: 'json'|'csv'|'markdown', conversation_id?: string, conversation_ids?: string[] }
- Export all or a subset filtered by
conversation_id
/conversation_ids
- Input:
- analyze_code_statistics
- Input:
{ days?: number, group_by?: 'day'|'week'|'month'|'language'|'workspace' }
- Aggregated metrics across the selected period
- Input:
- diagnose_storage
- Input:
{ limit?: number }
- Lists the largest
ItemTable
keys per workspace to help locate actual chat storage
- Input:
Ensuring assistant replies in results
- If your environment has a key containing full conversations (including assistant messages), the server will parse it directly.
- If not, the server falls back to
aiService.generations
(user-only prompts) and tries to pair them withaiService.prompts
as assistant replies heuristically. This improves completeness but may not perfectly reflect the original assistant responses. - To achieve exact assistant outputs, provide the confirmed storage key or path where Cursor persists assistant replies in your version.
Export formats
- JSON: raw conversations array
- CSV: one line per conversation with high-level metadata
- Markdown: per-conversation sections with messages and code blocks
Development
Notes
- SQLite access uses
sqlite3
. On some systems you may need build tools (e.g., Xcode CLT on macOS) installed. - Default workspace storage paths:
- macOS:
~/Library/Application Support/Cursor/User/workspaceStorage
- Windows:
%APPDATA%/Cursor/User/workspaceStorage
- Linux:
~/.config/Cursor/User/workspaceStorage
- macOS:
Troubleshooting
- No conversations returned:
- Use
diagnose_storage
to inspect keys. If no conversation-like key exists, your Cursor version may not store assistant replies locally. - The server will still return synthesized conversations from
aiService.generations
(user prompts). Provide the real chat key to enable full parsing.
- Use
- Permission/locking errors:
- Ensure Cursor is not locking the database when scanning. Try closing heavy operations in Cursor.
License
MIT — see LICENSE
.
This server cannot be installed
local-only server
The server can only run on the client's local machine because it depends on local resources.
Enables querying, searching, and analyzing Cursor IDE conversation history from SQLite workspaceStorage databases. Supports exporting chat data in multiple formats and provides workspace utilities for managing conversation data across projects.
Related MCP Servers
- -securityAlicense-qualityFacilitates integration with the Cursor code editor by enabling real-time code indexing, analysis, and bi-directional communication with Claude, supporting concurrent sessions and automatic reconnection.Last updated -2238TypeScriptMIT License
- -securityFlicense-qualityA Model Context Protocol server that enables AI assistants to explore and interact with Cursor IDE's SQLite databases, providing access to project data, chat history, and composer information.Last updated -21Python
- -securityFlicense-qualityA comprehensive memory management system for Cursor IDE that allows AI assistants to remember, recall, and manage information across conversations through a user-friendly interface.Last updated -2Python
- -securityFlicense-qualityA tool that provides simple API to execute SQL queries and manage MySQL databases, designed to integrate with Cursor IDE for AI assistants to directly perform database operations.Last updated -Python