Give Claude Code access to your issue tracker in 5 minutes (MCP)

Here is the whole thing in one line. Claude Code talks to external systems through MCP, the Model Context Protocol, so if your issue tracker ships an MCP server, you can connect it once and your agent will file, search, and close issues without you copying anything back and forth.

Most of the setup guides you will find stop at the filesystem and GitHub servers. This one is about pointing Claude Code at an issue tracker, which is where the move actually pays off: the agent stops asking you what to work on and starts reading the work itself.

#What MCP actually does here

MCP is an open standard. Claude Code is one client; an MCP server is the thing on the other end that exposes a set of tools. When you connect a server, its tools show up in Claude Code's tool list, and the agent can call them the same way it calls its built-in file and shell tools.

For an issue tracker, the tools are the obvious verbs: create an issue, search issues, list issues, comment, close an issue, run the triage queue. Once those are wired in, "what is left on RAD-219" and "file an issue for this flaky test" become things the agent does in its own loop, not things you go do in another tab and paste back.

Claude Code supports two transports: stdio for a local process, and HTTP for a remote, hosted server. The remote path is the shorter one, because there is nothing to install and authorization happens in your browser.

#The 5-minute setup

Radial ships a hosted MCP server at mcp.radial.build. Adding it to Claude Code is one command, then one approval. From your terminal:

claude mcp add --transport http radial https://mcp.radial.build

That registers the server. Now start Claude Code and run /mcp, which lists your connected servers and lets you authorize the new one. Radial uses OAuth, so the browser opens, you approve the connection against your workspace, and the token comes back to Claude Code. There is no API key to generate, copy, or paste, and you can revoke the connection later in one click without touching anything else.

If you would rather keep the config in your project so a teammate gets it on checkout, drop it in .mcp.json at the repo root instead of adding it by hand:

{
  "mcpServers": {
    "radial": {
      "url": "https://mcp.radial.build"
    }
  }
}

Either way, the next time the agent needs to know what to build, it asks the tracker instead of asking you.

#What your agent can do once it is connected

The server exposes the tracker's verbs as tools: create_issue, update_issue, search_issues, list_issues, comment, close_issue, list_projects, and triage_queue. So a session can go like this, in plain language to the agent:

• "Search Radial for open issues assigned to me in the ENG team, then start on the highest priority one."
• "I just fixed the login race condition. Close the issue and leave a comment with the root cause."
• "While you were working you found a second bug. File it as a new issue, priority high, before you move on."

The agent reads the work, does the work, and writes the result back to the same place. You are not the message bus between the model and the tracker anymore.

There is a second reason this matters, and it is the one developers keep running into on big projects. When a project gets large, the reasoning behind decisions scatters: some in a chat window that resets when it fills up, some in a doc that drifted out of date, some lost entirely. As one developer put it, GitHub stores the code but not why you wrote it that way. An issue the agent files and comments on as it works is the cheapest durable home for that reasoning, and it lives in the same place the next agent session will look.

#Local instead of remote

If you do not want a hosted connection, the same tools are available over stdio. The Radial CLI serves them locally:

radial mcp

Point any MCP client at that command and you get the identical tool set, running on your machine against your authenticated session. Same verbs, no remote server, your call.

#Where the line is

One honest note, because it is the whole point of the product. The intelligence here is your agent, your model, your keys. Radial is the fast system of record the agent writes to. There is no copilot, no AI summary, no auto-triage, and no AI credit meter inside the tracker, because the agent already does that part and you already pay for it once.

That is also why the connection does not cost you anything extra. Every agent credential is a client of the API, CLI, or MCP server, not a billed seat. The price is $50 per user, per year, flat, billed annually, for the humans. Agents ride free, and the Plain Software Pledge is the binding version of that: the day Radial ships a copilot, meters your usage, or charges you for AI you didn't ask for, your subscription is free.

#FAQ

#What is the difference between Claude Code and MCP?

Claude Code is Anthropic's command-line coding agent. MCP is the open protocol it uses to reach tools and data outside its built-in set. Claude Code is the client; an MCP server is what it connects to. They are not alternatives to each other; MCP is how Claude Code gets access to your issue tracker, your database, or anything else that speaks the protocol.

#How do I add an MCP server to Claude Code?

For a remote server, run claude mcp add --transport http <name> <url>, then run /mcp inside Claude Code to authorize it. For a local one, point a stdio config at the command that serves it. You can also check a .mcp.json file into your repo so the server is configured for everyone on the project.

#Do I need an API key to connect Claude Code to Radial?

No. The hosted server at mcp.radial.build uses OAuth, so you authorize the connection in your browser and Claude Code holds the token. Nothing to paste, and you can revoke it in one click. Scoped API keys exist if you want them for CI or a non-interactive agent, but they are not required for the Claude Code path.

#Does connecting my agent over MCP cost extra?

No. Agent credentials never count as billed seats. You pay $50 per user, per year, for the humans on the workspace, and that number is locked at the rate you join. There is no AI credit meter anywhere in the product.

#Can I use the same setup with Codex or Cursor?

Yes. The server is plain MCP, so any client that speaks the protocol can connect to the same eight tools. Claude Code is the example here because it is the most common, but the connection is not Claude-specific.

#The short version

Connecting Claude Code to your tracker takes one command and one browser approval, and after that the agent reads the work, does the work, and writes the result back to the place it already lives. That is the wedge: not a smarter tracker, a tracker your agent can drive.

See the full tool list and the OAuth flow on the developers page, or read why issue tracking isn't dead if you got here from the threads about whether the tracker still has a job.