langflow v1.13.1 — 1.13.1
- Source
- github.com
- Author
- github-actions[bot]
- Date

Context
Langflow 1.13.1 is a patch release made almost entirely of security and isolation fixes, and several of them harden the triggers system that 1.13.0 introduced, which lets a flow start on a schedule, a webhook or an event such as a Slack message. One fix fences stale listener delivery across trigger replicas (#15942). In systems that run several copies of a listener, fencing usually means stopping a copy whose claim on an event has lapsed from still delivering it; the title points to that problem but does not describe the mechanism. Slack workspace routing now requires OAuth attestation (#15981), listeners now stop when background consent is withdrawn (#16012), and SSO triggers are preserved during migration replay (#16058).
The largest remaining group concerns , the Model Context Protocol, a standard for connecting AI applications to outside tools. Fixes protect reserved fields in the MCP component, isolate login and request state in batch calls (#15978), validate shell arguments after headers are applied, enforce a destination policy for MCP HTTP requests (#16041), and pin agentic MCP calls to the backend listener. Other fixes confine the Git loader's files to its repository (#15972) and limit the Oracle loader's local file access, redact credentials from query strings in logs (#16042), separate user and flow storage namespaces, and secure the quickstart PostgreSQL configuration. Two more, which block runtime readers in generated components (#16040) and block re-export bypasses through inspect (#16048), appear to continue the checks on generated component code that 1.12.7 tightened.
These descriptions come from the pull request titles alone. The pull request bodies were not read, so this explanation cannot say how each fix works, how serious the underlying problems were, or whether any was reported or exploited. The release also upgrades dependencies, works around CVEs in the image's dependencies, and clarifies the automatic-login behavior that 1.13.0 turned off by default.
- Langflow 1.13.1 is mainly a hardening release: most of its listed bug fixes tighten file access, MCP calls, credentials and authentication rather than add features.
- File loaders are now confined: the Git loader keeps files inside the repository, and the Oracle loader's local file access is restricted.
- MCP handling got several fixes: reserved MCP component fields are protected, batch login and request state are isolated, MCP HTTP destinations follow a policy, and agentic MCP calls are pinned to the backend listener.
- Credential handling is tighter: credentials in query strings are redacted from logs, external resolver credentials are restricted, and unified model credential destinations are validated.
- Code execution paths are narrowed: untrusted global Python imports now require isolation, and fixes block runtime readers in generated components, inspect re-export bypasses, and assistant expression bypasses.
- MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
Checking sign-in…
Loading comments…






