The project system
your agent already speaks.
Hydrant lives where work happens — inside the conversation you're already having with your agent. File issues from a chat. Ship them from the same chat. The same actions are available in your shell.
brew install hydrant/tap/hydrantProject trackers were built for humans clicking buttons.
Your agent has codebase context. It knows the file the bug lives in, why the regression happened, what the fix probably touches. But until now, the only way to capture that into a ticket was for you to leave the conversation, open a tracker, and re-type what your agent already knew.
Hydrant exposes the same project actions over MCP and CLI — so "create the issue" and "ship the issue" stay inside the loop you and your agent are already in.
MCP or CLI.
Same project context underneath.
Some operators want agents to call tools directly. Others prefer commands they can inspect, paste, script, and run themselves. Hydrant keeps both surfaces on the same object model, permissions, identifiers, and audit trail.
Let the agent operate Hydrant directly.
Expose issues, decisions, milestones, notes, files, dependencies, and execution briefs as structured tools inside any MCP-capable agent.
get_issuecreate_issueset_dependenciescreate_decisionlist_decisionsupdate_decisioncreate_milestonelist_cyclesget_cycle_issuescreate_notelist_noteslist_folderssave_filelist_filesget_file_urlsearchpreview_bundleand_more{
"mcpServers": {
"hydrant": {
"type": "http",
"url": "https://hydrant.dev/mcp",
"headers": { "Authorization": "Bearer hyd_sk_…" }
}
}
}Run the same actions from the terminal.
Pipe-friendly commands for humans and agents running in shell mode. Same backend, same IDs, same dependency graph — no context fork.
brew install hydrant/tap/hydrant hydrant auth login --pat hyd_sk_… hydrant use hydrant
When MCP and CLI aren't where the work is.
Hydrant's core is the object model: issues, decisions, notes, dependencies, files, milestones, and audit. These secondary surfaces exist for moments where the right interface is a webhook, service, CI runner, or borrowed shell.
For automation and CI.
Plain JSON over HTTPS for webhooks, CI runners, and services that should not install a binary or host an MCP client.
{
"title": "Audit logout on stale tokens",
"priority": "high",
"blocks": ["HYD-489"],
"labels": ["auth", "security"]
}For zero-install moments.
A temporary terminal view for borrowed laptops, fresh containers, and shells where installing a binary is the wrong move.
For reading the room, not pushing the buttons.
The Feed is the cockpit: what changed, who changed it, which issues heated up, and what your agents just did. It is built for review and orientation when the terminal scroll is no longer enough.
- Grouped by issue so agent activity is readable, not noisy.
- Insight ribbon pulls out blockers, bursts of work, and review-ready threads.
- Filters split humans and agents without forking the underlying history.
Activity