Connect Claude Code to your backlog

Why your AI agent needs your backlog

A coding agent is only as good as the context it can see. Point it at the repo alone and it writes plausible code for the wrong requirement. Stride gives the agent the structured context it’s missing (acceptance criteria, the linked test cases, the architecture decision that constrains the change, the sibling stories in the sprint) so it ships the feature you actually scoped, not just code that compiles.

How to connect it

1. In Stride, open Settings → Integrations → Claude Code, pick the project the agent should reach, and click Generate Agent Key.
2. Put the key in an environment variable: export STRIDE_AGENT_KEY="nl_your_agent_key" on macOS and Linux, or setx STRIDE_AGENT_KEY "nl_your_agent_key" on Windows, then open a new terminal. The config below refers to the variable, so the key never sits in a file you might commit.
3. Paste the config into .mcp.json at your repository root to share the server with the project. Claude Code asks you to approve a project server the first time it loads it.

{
  "mcpServers": {
    "stride": {
      "type": "http",
      "url": "https://www.stride.page/api/mcp/stride",
      "headers": {
        "Authorization": "Bearer ${STRIDE_AGENT_KEY}"
      }
    }
  }
}

To use Stride in every project instead, add it once from your terminal: claude mcp add --transport http stride --scope user https://www.stride.page/api/mcp/stride --header "Authorization: Bearer $STRIDE_AGENT_KEY" (in PowerShell, write $env:STRIDE_AGENT_KEY). The shell fills in the key, so Claude Code stores it in ~/.claude.json.

Then run claude mcp list: Stride should show as connected, and /mcp inside Claude Code shows its tool count. Claude Code discovers Stride’s tools automatically over MCP. There’s no plugin to install. Codex and Cursor work the same way — switch tabs above for the same endpoint in their own format.

What the agent can do

Stride exposes 131 RBAC-gated tools across all four modules. Every call is permission-checked by the dispatcher before it runs, and every write lands in the audit log as via: "mcp".

Good to know

• 131 tools don’t crowd the context. Claude Code’s tool search is on by default: a session starts with only the tool names, and a tool’s full definition loads when Claude needs it. Claude Code sets no per-server tool limit.
• Pre-approve the tools you trust. Claude Code asks before using an MCP tool. Stride’s tools appear as mcp__stride__stride_<tool>, so the permission rule mcp__stride__* allows them all, while mcp__stride__stride_get_* and mcp__stride__stride_list_* allow only tools that read.
• If Stride shows as failed, the key did not reach it. Usually STRIDE_AGENT_KEY was not set in the terminal Claude Code started from, and claude mcp list warns about the missing variable. Set it and restart Claude Code.
• No timeout to configure. Claude Code gives each request to a remote MCP server 60 seconds by default. Stride answers within that, and long generations such as a diagram carry on in the background, where the agent can check on them.

Checked against Claude Code’s documentation and release notes on October 20, 2026 (version 2.1.305).

Frequently asked questions