---
type: "post"
id: "deployhq-launches-a-hosted-read-only-mcp-server-715662"
slug: "deployhq-launches-a-hosted-read-only-mcp-server-715662"
title: "DeployHQ launches a hosted, read-only MCP server for Claude Code, Codex, Gemini CLI and Cursor"
description: "DeployHQ has launched a hosted Model Context Protocol endpoint for asking AI assistants about deployment state."
retrieval_nugget: "DeployHQ has launched a hosted Model Context Protocol endpoint for asking AI assistants about deployment state. It removes the local package and stored API key required by its earlier MCP package, while starting with a deliberately read-only permission set for production-facing information."
published_at: "2026-10-30"
updated_at: "2026-10-30"
record_date: "2026-10-30"
date_kind: "published_at"
topics: ["mcp","claude","claude-code"]
entities: ["DeployHQ","DeployHQ Hosted MCP Server","Model Context Protocol","Claude Code","Codex","Gemini CLI","Cursor"]
related_patterns: ["agent-protocols-become-interoperability-layer","agent-ready-software-exposes-capabilities","agent-security-moves-to-runtime-boundaries"]
novelty_label: "new_fact"
what_changed: "DeployHQ launches its own hosted, read-only MCP server for querying deployment status, a distinct product release from the prior MCP servers."
prior_refs: ["memory:ccf2ac23-3ec9-8098-a2b9-78f35eca552a","memory:51f879dc-1053-8a4d-aba0-15995253df5a"]
source_url: "https://www.deployhq.com/blog/introducing-the-deployhq-hosted-mcp-server"
source_urls: ["https://www.deployhq.com/blog/introducing-the-deployhq-hosted-mcp-server"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:deployhq-launches-a-hosted-read-only-mcp-server-715662"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/deployhq-launches-a-hosted-read-only-mcp-server-715662/","markdown":"https://newruntime.com/posts/deployhq-launches-a-hosted-read-only-mcp-server-715662.md","json":"https://newruntime.com/posts/deployhq-launches-a-hosted-read-only-mcp-server-715662.json"}
---

# DeployHQ launches a hosted, read-only MCP server for Claude Code, Codex, Gemini CLI and Cursor

## Retrieval answer

DeployHQ has launched a hosted Model Context Protocol endpoint for asking AI assistants about deployment state. It removes the local package and stored API key required by its earlier MCP package, while starting with a deliberately read-only permission set for production-facing information.

## Summary

DeployHQ has launched a hosted Model Context Protocol endpoint for asking AI assistants about deployment state. It removes the local package and stored API key required by its earlier MCP package, while starting with a deliberately read-only permission set for production-facing information.

## What happened

DeployHQ announced its hosted MCP server on 30 October 2026. The endpoint is hosted at `mcp.deployhq.com/mcp` and is intended to let an AI assistant query DeployHQ accounts, projects, servers, and deployments in plain language. The company positions it as a replacement for the setup burden of its earlier local MCP package, which required installing a package, creating an API key, and retaining that key in a local configuration file.

The new setup is hosted rather than locally run: a user adds the endpoint address to an assistant, signs in to DeployHQ through a browser, and approves the connection. DeployHQ says the connection acts as that user, so it can access only what that user can already access in DeployHQ. The post provides terminal setup commands for Claude Code and Codex, and says Gemini CLI, Cursor, VS Code, Claude.ai, and ChatGPT have step-by-step setup coverage elsewhere in its documentation.

At launch, the server exposes six read tools: `list_accounts`, `list_projects`, `get_project`, `list_servers`, `list_deployments`, and `get_deployment`. These cover project settings such as automatic deployments and repository branch; server information including protocol, environment, server group, and auto-deploy status; and deployment details including status, revisions, timings, servers, and build settings. DeployHQ gives examples such as finding projects that have gone longest without a deployment, checking recent failed deployments, identifying servers used by a project, and inspecting the branch and revisions of a latest deployment.

The boundary is explicit. The hosted server cannot start, retry, or roll back a deployment, change settings, read deployment logs, or reveal server passwords, keys, host names, paths, environment variables, configuration-file contents, billing data, or account details. Repository addresses are stripped of embedded user names and passwords. Lists are also bounded: 200 projects, 200 servers per project, and 500 recent deployments per project, with truncation signalled to the assistant. Account administrators can disable assistant access across an account immediately. A plan with API access is required; single-sign-on accounts cannot yet use browser-signing assistants such as Claude.ai and ChatGPT, though a local assistant can use an API key. DeployHQ says write tools and broader browser-assistant support are planned, but gives no delivery date.

## What the archive adds

This is a new event in a related story: vendors are exposing operational systems to agents through MCP. [Fly.io’s MCP server](https://newruntime.com/posts/fly-io-launches-the-fly-mcp-server-mcp-a5495a/) similarly connected agents to infrastructure resources, but DeployHQ’s launch is specifically about deployment visibility and begins with a hosted, read-only model.

## Trend signal

The release supports the assumption that developers will use multiple AI clients over the same projects, rather than converge on one vendor interface: DeployHQ documents connections across Claude Code, Codex, Cursor, Gemini CLI, and browser assistants. It also supports the related expectation that MCP can provide a shared external context layer across clients: [shared context infrastructure](https://newruntime.com/shifts/shared-context-infrastructure/). More broadly, it fits [agent protocols become the interoperability layer](https://newruntime.com/patterns/agent-protocols-become-interoperability-layer/): the same deployment capability is exposed through a standard endpoint rather than a single assistant-specific integration.

## Why now

DeployHQ frames the launch as a simplification of its own earlier local MCP package. The prior approach required a local install and an API key held in configuration; the hosted endpoint moves connection and approval into a browser sign-in flow. Its read-only launch also lets the company expose useful operational context before it exposes commands that can alter production deployments.

## Connections to other events

Fly.io’s recent MCP server connected agents to Sprites and Fly Machines through a similar protocol surface. The distinction is useful: Fly.io exposed infrastructure resources, while DeployHQ exposes deployment records and server configuration. Both illustrate MCP becoming a common way for an assistant to retrieve tool-specific operational context without requiring the assistant vendor to build a bespoke integration.

## Why it matters

For teams already using DeployHQ, deployment status becomes queryable inside the coding environment where an incident, pull request, or release is being discussed. That can remove dashboard navigation for routine checks, especially across many projects or client accounts. The practical value is not autonomous deployment: it is faster retrieval of bounded operational facts—what deployed, where it deployed, whether it failed, and which revision was involved.

The more important design choice is the permission model. DeployHQ separates read access from write actions, excludes secrets and logs, scopes visibility to the authenticated user, and gives administrators an account-level off switch. Teams evaluating MCP connections to production systems should treat that shape as a review baseline, not assume that a conversational interface makes deployment control safe by default.

## Limits and uncertainty

The announcement is a vendor description of the service at launch, not an independent security assessment. It does not establish how reliable assistant interpretation of the returned deployment data will be, how organizations should audit individual assistant queries, or when write tools will become available. It also does not say whether the current caps suit larger deployment estates. Read-only does not mean risk-free: project, server, branch, revision, and timing data can still be sensitive operational context, and the post leaves each team to decide which users should be allowed to expose it to an assistant.

## Skill to build now

Create a short approval checklist for any MCP connection to operational systems. Record the authenticated identity, tools exposed, data excluded, list limits, revocation path, and whether the server can write. Then test three real read-only questions—recent failures, latest revision, and target servers—against the dashboard before relying on assistant answers during a release. This turns a convenient integration into a verifiable operational aid.

## Reader takeaway

DeployHQ’s hosted MCP server is not a deployment agent yet. It is a browser-authorized, user-scoped read layer for deployment context across several AI clients. Its value is quicker operational lookup; its notable lesson is that production MCP integrations can begin with narrow, reversible, observable permissions rather than immediate write access.

## Open question

Will DeployHQ retain separate approval and verification controls when it introduces write tools, or will deployment actions be added to the same connection scope?

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Mcp](https://newruntime.com/topics/mcp.md): Explore the mcp topic hub.
- topic hub: [Claude](https://newruntime.com/topics/claude.md): Explore the claude topic hub.
- topic hub: [Claude Code](https://newruntime.com/topics/claude-code.md): Explore the claude-code topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)