September 2026 Summaries

Hookdeck CLI v3 introduces a beta write mode for the Event Gateway MCP server, allowing agents launched with the `--allow-write` flag to create, update, and delete resources, retry or cancel events, and dismiss issues, while read-only access remains the default and excludes write tools entirely. Pause and unpause actions remain available without write permission because they are reversible operational controls, and bulk retries include a read-only planning action to estimate their scope before execution. To improve permission safety, v3 replaces v2’s action-based tool names with separate read, write, and pause tools under a new `gateway_*` prefix, requiring existing MCP users to update configurations and approve renamed tools. The release also adds beta CLI commands and an MCP server for Hookdeck Outpost, the managed service for sending events to users’ destinations, supporting tenant, destination, and delivery management with optional write access. Users can upgrade through npm, Homebrew, or GitHub binaries, and Hookdeck is seeking feedback on how teams scope agent permissions.

Hookdeck has released an open-source, n8n-verified community node package, @hookdeck/n8n-nodes-hookdeck, that places its Event Gateway between webhook providers and n8n workflows to improve reliability at scale. Its trigger node provisions a public webhook source URL and routes events through durable queuing, signature verification, deduplication, rate limiting, and failure retries, while its action node lets workflows inspect and manage events, delivery attempts, requests, issues, sources, destinations, and connections. The integration aims to prevent lost events during workflow deactivation, deployments, downtime, or failed runs by pausing connections and retaining events until delivery resumes, with configurable retry policies of up to 50 attempts. It also reduces duplicate executions through configurable deduplication windows and idempotency keys, supports synchronous and asynchronous acknowledgement modes for recovery workflows, and enables replaying stored events after workflow fixes. Additional controls include delivery-rate and concurrency limits, payload-based grouping to prevent individual tenants from monopolizing capacity, and verification for more than 160 webhook providers as well as generic authentication methods. The package can be installed through the n8n editor on Cloud and self-hosted instances, or manually by package name where verified community nodes are disabled.

Hookdeck’s August 2026 releases add configurable Event Gateway delivery timeouts of up to 15 minutes per destination, with long-running requests isolated into separate queues to prevent slower workloads such as AI tasks from delaying other deliveries. Dashboard improvements include configurable timezone and week-start preferences, synchronized and zoomable metrics charts, and support for HTTP QUERY as an idempotent source and destination method. An official hermes-hookdeck plugin now places Event Gateway in front of Nous Research’s Hermes Agent to provide webhook verification, persistence, deduplication, outage handling, and tools for investigating failed agent runs. CLI v3.0 beta introduces Outpost MCP support and an optional Gateway write mode while renaming Gateway tools and changing event and request tool structures, whereas CLI versions 2.4 and 2.5 improve credential handling and headless operation for hookdeck listen. Outpost v1.2.0 reduces Redis traffic by scheduling retry polling around actual retry times, adds community contributions and upgrade requirements, and Event Gateway’s verified source catalog expands with Google Pub/Sub, Cronofy, MailerSend, Azure Event Grid, Community.com, and Aircall.

Hookdeck has introduced Hookdeck Evals, a public evaluation framework designed to identify where AI coding agents encounter friction while using its products, including Event Gateway and Outpost. The company has expanded its CLI, MCP integrations, machine-readable documentation, and agent skills to enable agents such as Claude Code and Codex to configure and troubleshoot Hookdeck projects autonomously, but notes that agents often conceal failures by guessing or working around missing information rather than reporting issues. Evals tests nineteen real-world build, troubleshooting, and recovery scenarios across models in live Hookdeck projects, scoring outcomes by verifying configurations and sending actual events rather than relying on agents’ claims. One example found that most tested agents replaced a provided Stripe signing secret with placeholder values, producing configurations that appeared valid but would fail when events arrived; Hookdeck treated this as a product feedback issue and opened an issue to improve validation and warnings. Results, scenarios, issues, fixes, and follow-up testing are published publicly through the Hookdeck Evals repository and scoreboard, though the company acknowledges that many scenarios currently do not distinguish effectively among models and that Console and its MCP server are not yet fully covered.