Why this exists
A coding agent that cannot read your requirements is guessing. It re-proposes solutions you rejected last week, invents acceptance criteria, and writes tests for behaviour nobody asked for — because every session starts from nothing.
Oprex exposes the whole lifecycle over the Model Context Protocol. Your agent reads the requirement, checks which tests cover it, sees the bug it is fixing and the release it is going into, and writes its findings back where the next session — human or machine — will find them.
Connect Claude Code
Create a key at Settings → API Keys. The install command, with your key already in it, is shown on screen the moment the key is created — that is the only time the key exists in plain text.
claude mcp add --transport http oprex https://oprex.id/mcp \
--header "Authorization: Bearer oprex_pk_…"
Then claude mcp list should show oprex. That is the whole setup.
Other clients
Claude Desktop
Uses stdio, so it needs the mcp-remote bridge. The exact config is on the setup page.
Cursor, Antigravity, and others
Anything speaking the standard mcpServers schema works — HTTP transport, one Authorization header.
OAuth clients
Clients that prefer a consent flow over a static key can register dynamically and obtain a token with the scopes you approve.
What the agent can do
| Capability | Examples |
|---|---|
| Read the lifecycle | Projects, requirements, specifications, test cases, milestones, releases, release plans, coverage. |
| Work defects | List, read, create, and update bugs and helpdesk tickets, with comments. |
| Keep memory | Create and update long-form Memory documents and short Notes, so the next session inherits context instead of rebuilding it. |
| Resolve names | Turn a human path like acme/checkout into the ids the other tools need. |
Safety
- Scoped keys. Give the agent a read-only key while you evaluate it. Write methods are refused at the auth layer, not by convention.
- Narrow it to one project. A key can be bound to a single group or project, so an agent working on one thing cannot read the rest.
- It acts as you. Per-project access control applies to the agent exactly as it applies to you — connecting an agent never widens what is visible.
- Revocable. Keys can be disabled or deleted at any time, and every key shows when it was last used.
Give your agent a memory
Create a workspace, create a key, paste one command. The whole thing takes about two minutes.
Create an MCP key Ask the community