cursor-chat-browser

Cursor Chat Browser

Search and retrieve past Cursor AI conversations across all workspaces using the cursor-chat-browser MCP server.

Prerequisites

The cursor-chat-browser MCP server must be configured in ~/.cursor/mcp.json. It indexes conversations from ~/.cursor/projects/*/agent-transcripts/ into a SQLite FTS5 search index at ~/.cursor/chat-browser/search-index.db.

Operations

Search for past conversations

Use the search_conversations MCP tool to find conversations by keyword, function name, or concept:

MCP tool: cursor-chat-browser -> search_conversations
Arguments:
  query: "detection rules migration"     # keywords, function names, concepts
  workspace: "kibana"                     # optional: filter by project (partial match)
  limit: 10                               # optional: max results (default: 10)

The search uses full-text search with Porter stemming, so "migrating" matches "migration", "detect" matches "detection", etc.

Retrieve a full conversation

After finding a relevant conversation via search, retrieve its full content:

MCP tool: cursor-chat-browser -> get_conversation
Arguments:
  id: "uuid-from-search-results"          # conversation UUID
  max_length: 10000                        # optional: truncate long conversations (default: 10000)

Browse recent conversations

Get the most recent conversations, optionally filtered by workspace:

MCP tool: cursor-chat-browser -> recent_conversations
Arguments:
  workspace: "elastic-cursor-plugin"       # optional: filter by project
  limit: 10                                # optional: number of results (default: 10)

List all workspaces

Discover which projects have indexed conversations:

MCP tool: cursor-chat-browser -> list_workspaces

Reindex new conversations

If new conversations were created since the server started:

MCP tool: cursor-chat-browser -> reindex

When to Use

Scenario	What to Do
Starting a complex feature	Search for prior discussions about the same area
User says "we discussed this before"	Search for the referenced topic
Debugging a recurring issue	Search for past debugging sessions
Making an architectural decision	Search for previous architecture conversations
Working in unfamiliar codebase area	Browse recent conversations for that workspace
Onboarding to a project	List workspaces, then browse recent conversations

Guidelines

• Search first, then retrieve: Use search_conversations to find candidates, then get_conversation only for the most relevant 1-2 results. This avoids context bloat.
• Be specific with queries: "elasticsearch authentication API key rotation" is better than "auth".
• Use workspace filters: When you know the project, filter by workspace name to reduce noise.
• Respect context limits: Set max_length appropriately -- 10,000 chars is usually enough to understand the discussion without overwhelming the context window.
• Skip for trivial tasks: Don't search for past conversations when doing typo fixes, config changes, or single-file edits.
• Combine with other context: Past conversations complement -- don't replace -- reading the actual code, git history, and documentation.

Examples

"What approach did we use for the Cypress to Scout migration?"

search_conversations(query: "cypress scout migration approach", workspace: "kibana")

Then retrieve the top result:

get_conversation(id: "<result-id>", max_length: 15000)

"What have I been working on recently?"

recent_conversations(workspace: "kibana", limit: 10)

"Find where we discussed the skill authoring workflow"

search_conversations(query: "skill authoring workflow")

"How many projects have conversation history?"

list_workspaces()