Skip to content
You are reading the docs for Backstory 0.2 (private beta). Behavior may change before general availability; the changelog lists every change.

MCP server

The Backstory MCP server speaks Streamable HTTP at https://api.backstory.io/mcp and authenticates with a scoped API token. Create a token with the mcp scope under Organization → API tokens. PHI organizations can disable the server or restrict it to non-PHI projects.

ToolWhat it returns
list_recent_issuesOpen issue clusters ranked by impact, with counts and releases
get_session_transcriptThe Semantic Transcript for a session (masked, cited)
get_session_summaryThe automatic summary artifact when one exists
get_error_contextSymbolicated stack, request, and surrounding steps for an error
get_repro_stepsRepro steps and the generated Playwright test (creates a deep analysis if none is cached)
get_dom_snapshot_atThe reconstructed DOM at a moment, as sanitized HTML
get_request_detailsRequest metadata, timing, and linked backend trace
search_sessionsNatural-language search over sessions
get_issue_clusterCluster detail with samples and the AI explanation

Claude Code (~/.claude/settings.json or project .mcp.json):

{
"mcpServers": {
"backstory": {
"type": "http",
"url": "https://api.backstory.io/mcp",
"headers": { "Authorization": "Bearer backstory_…" }
}
}
}

Cursor and other clients accept the same URL and header.

“Fix the upload 413 issue from Backstory.”

The agent calls list_recent_issues, picks the cluster, calls get_repro_steps and get_error_context, edits upload.ts, runs the generated test, and opens a PR. Every tool response includes citations back to replay moments so a human can verify.

The MCP server ships with the AI phase. Tokens with the mcp scope can be created now; the endpoint returns 503 until the feature is enabled for your organization.