What the Hydrant MCP server does
The Hydrant MCP server gives an AI agent access to a Hydrant workspace for issue and project tracking. It can read workspace context and, when the authenticated user has the required permission, create or update issues, add comments, and manage relationships such as dependencies. Typical requests include listing open issues and blockers, summarizing activity on an issue, or filing a new issue from a bug report.
The repository also contains optional agent skills that guide ticket lifecycle work. These skills can help an agent capture and refine issues, prepare assigned work, build changes on a branch, handle review comments, and ship an issue under an explicitly confirmed grant. The skills are guidance for agent behavior; the MCP server supplies the underlying tools.
How it works
Hydrant MCP server connections use the hosted endpoint at https://hydrant.dev/api/mcp with the Streamable HTTP transport. The repository is not the hosted application's source, so cloning it is not required and does not create a local server.
After adding the endpoint to an MCP-compatible client, authenticate through that client's browser OAuth flow. Sign in with GitHub, select the intended workspace, and approve the access requested by the connection. No static API key is required for this OAuth flow. Permissions remain bounded by the user's workspace role and can be revoked from Hydrant under Settings, Agents, and Access.
Confirm the connected workspace before asking the agent to make changes. Write operations require permitted access; read-only requests can be used first to identify the workspace and inspect current issues or activity.
Setup and configuration
Add the remote server to the client's MCP configuration using an HTTP entry such as:
{
"mcpServers": {
"hydrant": {
"type": "http",
"url": "https://hydrant.dev/api/mcp"
}
}
}
Exact configuration syntax varies by client, so use the client's remote MCP setup process and Hydrant's connection guidance. The connection itself does not require an environment variable or a locally generated key.
The optional skills are installed separately with the skills CLI. For Claude Code and Codex, the repository documents npx skills add Background-Craft/hydrant-skills -s hydrant-setup -a claude-code -a codex, followed by asking the agent to run hydrant-setup. The setup skill scans repository instructions, scripts, CI, deployment configuration, and existing skills before presenting a plan. It does not connect the client, create credentials, or edit AGENTS.md or CLAUDE.md.
Tools and capabilities
Hydrant MCP server capabilities supported by the provided material include:
- Reading Hydrant workspace context.
- Listing or inspecting issues and their activity.
- Creating and updating issues when write access is allowed.
- Adding issue comments.
- Managing issue relationships and dependencies.
- Supporting optional workflow guidance for capture, refinement, preparation, implementation, review triage, and shipping.
The workflow skills also describe checks such as duplicate detection, acceptance criteria, repository profiles, CI gates, and review handoff. Their behavior depends on the repository profile and the grants or permissions available to the agent.
Limitations and notes
The hosted server's application source is private. This public repository contains skills and client integration packages rather than a local server implementation. The documented connection therefore depends on Hydrant's hosted service.
Write actions are permission-dependent, and access is constrained by the workspace role. The skills recommend confirming the workspace before changes and stopping when required clearance or review conditions are missing. The go workflow does not merge changes, while ship requires a confirmed grant for the specific issue.
Installing the skills does not authorize an account or connect an MCP client. The skills CLI writes skill files and a lockfile; it does not create keys or change Hydrant permissions. OAuth access can be revoked in Hydrant.