AI Sessions

An MCP server that makes sessions from Claude Code, OpenAI Codex, Gemini CLI, opencode, Mistral Vibe, and GitHub Copilot CLI available to any MCP-compatible client.

Allows AI agents to search, list, and read your previous local coding sessions from multiple CLI coding agents. Useful for:

- Finding past solutions to similar problems
- Reviewing what you worked on recently
- Learning from previous conversations
- Resuming interrupted work

Resuming a Claude Code session in Codex CLI.

macOS (Intel and Apple Silicon), Linux (amd64 and arm64, including WSL), and Windows amd64 (Git Bash):

curl -fsSL https://aisessions.dev/install.sh | bash

This installs the binary to~/.aisessions/bin. Follow the instructions to add it to your PATH.

curl -fsSL https://aisessions.dev/install.sh | INSTALL_DIR=/custom/path bash

Download pre-built binaries fromGitHub Releases.

go build -o bin/aisessions ./cmd/ai-sessions

claude mcp add --scope user --transport stdio ai-sessions -- ~/.aisessions/bin/aisessions

claude mcp add --scope user --transport stdio ai-sessions -- /path/to/aisessions

codex mcp add ai-sessions -- ~/.aisessions/bin/aisessions

[mcp_servers.ai_sessions] command = "/Users/YOUR_USERNAME/.aisessions/bin/aisessions"

ReplaceYOUR_USERNAMEwith your actual username, or use your custom install path.

Add to the config file opened fromSettings→Developer→Edit Config:

{ "mcpServers": { "ai-sessions": { "command": "/Users/YOUR_USERNAME/.aisessions/bin/aisessions" } } }

ReplaceYOUR_USERNAMEwith your actual username, or use your custom install path.

Restart Claude Desktop, then openDevelopersettings or+→Connectorsin a conversation to check the connection. Claude Desktop also supports packaged.mcpbextensions, but direct configuration remains useful for a standalone downloaded binary.

Theaisessionsbinary includes a CLI tool for uploading supported local agent transcripts toaisessions.devfor sharing.

aisessions upload /path/to/session.jsonl aisessions upload /path/to/session.jsonl --title "Custom Title"

- "Let's continue my latest session from Claude Code"
- "Show me my recent Codex sessions"
- "Search my sessions for authentication bugs"
- "How many times did Claude tell me I wasabsolutely rightyesterday?"

- Claude Code:~/.claude/projects/[PROJECT_DIR]/.jsonl
- Gemini CLI:~/.gemini/tmp/[PROJECT_HASH]/chats/session-.jsonl(plus legacy.jsonrecordings)
- OpenAI Codex:~/.codex/sessions/and~/.codex/archived_sessions/
- opencode:~/.local/share/opencode/opencode.db(plus the legacystorage/JSON tree)
- Mistral Vibe:~/.vibe/logs/session/
- GitHub Copilot CLI:~/.copilot/session-state/[SESSION_ID]/events.jsonl(plus legacy flat JSONL files)

When you ask your AI agent to list or search sessions, the server reads these local stores through source-specific adapters; it does not launch the agent CLIs.

Shows which AI CLI source adapters are available from the running server.

Lists recent sessions from all projects (newest first).

- source(optional): Filter byclaude,gemini,codex,opencode,mistral, orcopilot
- project_path(optional): Filter by specific project directory
- limit(optional): Max results (default: 10)

Example:{"source": "claude", "limit": 20}

Searches session content using BM25 ranking. Returns results sorted by relevance score with contextual snippets.

- query(required): Search term (supports multiple keywords)
- source(optional): Filter by source
- project_path(optional): Filter by project
- limit(optional): Max results (default: 10)

Example:{"query": "authentication bug"}

- session: Session metadata (ID, source, project, timestamp)
- score: Relevance score (higher = more relevant)
- snippet: Contextual excerpt (~300 chars) showing where the match occurred

Retrieves full session content with pagination.

- session_id(required): Session ID from list results
- source(required): Which coding agent created it
- page(optional): Page number (default: 0)
- page_size(optional): Messages per page (default: 20)

To keep formatting consistent and catch regressions early:

- Installpre-commitand runpre-commit installto enable hooks (gofmt,go vet,go test).
- Pushes tomainand pull requests targetingmainrun the GitHub Actions workflow (.github/workflows/build.yml), which checks formatting, runsgo vet, builds the binary, and executesgo test -cover ./....

Semantic code search for AI agents without indexing your codebase or storing any data. Fast and accurate.