RelayHistory MCP Server: Setup for Claude Code, Cursor and Codex

• Searches every past coding-agent session on your machine by keyword

  Looks through your saved prompts and the assistant text, tool calls and results from past sessions, newest first. Plain words must all match. You can narrow by tool, project, tag, role or date.

  Try: “Find the session where I rewrote the auth flow, and show the newest matches first.”

  Read · search_history

• Lists your most recent prompts across all tools, newest first

  Returns the latest prompts already saved in the index. You can filter by tool, project, tag or date range and page backwards.

  Try: “Show my last 20 prompts across Claude Code and Codex.”

  Read · recent_history

• Lists the sessions in your local index

  Shows the catalog of sessions already indexed, filtered by tool. It never scans for new sessions or runs a sync.

  Try: “List my ten most recent sessions in Cursor.”

  Read · list_sessions

• Gets the prompts of one session

  Returns the prompts saved for a single session. Pass a tool name when two tools reuse the same session id.

  Try: “Get the prompts from that auth rewrite session.”

  Read · get_session

• Reads one page of a session's step-by-step events

  Returns a page of normalized events for a session, such as assistant messages and tool activity. You continue from the last row to get the next page.

  Try: “Show the first 50 events of that session.”

  Read · get_session_events

• Lists the tool calls a past session made

  Returns one page of the tool calls recorded for a session, so you can see what the agent ran.

  Try: “Which tools did my last Codex session call?”

  Read · get_session_tool_calls

• Lists the files a past session edited

  Returns one page of the file edits recorded for a session.

  Try: “Which files did yesterday's Claude Code session change?”

  Read · get_session_file_edits

• Shows the sub-agents a session started, as a tree

  Builds the full tree of sessions that grew out of one session, following delegation by default. You can also include resumed, continued and forked sessions.

  Try: “Show the sub-agent tree for this session.”

  Read · get_session_tree

• Shows which sessions a session spawned or came from

  Lists direct links for one session: sessions it delegated to, the session that started it, and resume, fork and continuation links.

  Try: “Which session was this one forked from?”

  Read · get_session_relationships

• Lists non-chat records such as compaction points

  Returns a page of records a tool wrote that are not chat turns, such as compaction and summary boundaries, system rows and agent start and stop events.

  Try: “Where did this session get compacted?”

  Read · get_session_markers

• Lists the model requests of a session with token usage

  Returns one row per model request with usage normalized, so the numbers can be summed. Cost is shown only when the source data carried one.

  Try: “How many requests did that session make, and how many tokens did each use?”

  Read · get_session_requests

• Totals the tokens one session used

  Rolls up token usage for a session across tools. Missing usage is reported as missing, not as zero, and cost appears only if the source recorded it.

  Try: “How many tokens did my last Codex session use?”

  Read · get_session_usage

• Shows how much history is indexed

  Returns statistics about the history already in your index, optionally for one tag.

  Try: “How much history do you have indexed, by tool?”

  Read · history_stats

• Says what one tool's history can and cannot tell you

  Reports which kinds of evidence RelayHistory can record for one tool, and what its records show about delegation. It answers before any sync.

  Try: “What can you record from Cursor sessions?”

  Read · get_source_capabilities

• Re-reads all your tools' session files into the index

  Runs a full ingestion of session data from every supported tool into the local index. With a remote scope and a configured source plugin it can also reach provider services.

  Try: “Sync my agent history so search is up to date.”

  Write · sync

• Scans for new sessions and adds them to the catalog

  Finds sessions quickly and updates only the session catalog, without indexing their full content. A remote scope needs a configured source plugin.

  Try: “Check for new sessions from Codex and Cursor.”

  Write · discover_sessions

• Indexes one session in full

  Loads a single cataloged session as deeply as its tool allows, without a global sync. Tools that record only prompts come back marked partial.

  Try: “Index the full content of that Codex session.”

  Write · hydrate_session

• Makes a pointer so another agent can pick up this session

  Builds a handoff pointer for your current session. Its intent field is a ready-made prompt that tells the receiving agent to resume the session and carry on. It may add the session to the local catalog.

  Try: “Create a handoff for this session so a teammate's agent can finish the migration.”

  Write · create_handoff

• Loads a handed-off session so you can continue it

  Fetches a session from a handoff pointer in your own workspace and returns its prompts, events, tool calls and file edits. Handoffs from other workspaces or organizations are rejected.

  Try: “Resume the handoff my teammate sent and continue the work.”

  Write · resume_handoff

• Lists agents that are live on Agent Relay

  Shows live participants on Agent Relay, not session history. It reads only the local Agent Relay desktop app and needs no cloud credential.

  Try: “Who is on Agent Relay right now?”

  Read · list_relay_agents

• Checks whether this session is reachable on Agent Relay

  Reports whether the session running this server is registered and reachable on Agent Relay.

  Try: “Am I reachable on Agent Relay?”

  Read · relay_status

• Makes this session reachable by teammates and agents

  Puts the session running this server on Agent Relay. The optional name and description you give are public relay metadata.

  Try: “Join Agent Relay as 'billing-refactor'.”

  Write · join_relay

• Takes this session off Agent Relay

  Removes the session running this server from Agent Relay.

  Try: “Leave Agent Relay.”

  Write · leave_relay

Watch out for

• Everything you typed in past sessions is searchable

  The index holds prompts, assistant text, tool calls and tool results from every supported tool. Anything pasted into an old session, such as a key, can come back in a search result and goes to your assistant's model provider. github.com

• join_relay shows your session to teammates and agents

  It puts your running session on Agent Relay, and the name and description you give are public. Use leave_relay or relay_status to undo and check it; the Agent Relay desktop app is required. github.com

• sync and hydrate write to the local index

  These tools read the session files of your other tools and store them in a SQLite database on your machine. With a remote scope and a source plugin they can also contact provider services. github.com

• Remote connectors reuse existing sign-ins

  The optional claude-web connector reads the Claude Code CLI's stored OAuth token. It only runs if you install and configure the provider-sources plugin. github.com

• A handoff loads another agent's session into your chat

  resume_handoff returns the full prompts and tool output of a session from your workspace. Our advice: treat that content as untrusted text, as you would any pasted document. github.com

Security advisories