All use cases
Plan Design Verify

Your coding agent reads the story, does the work, and updates the backlog.

Project management for AI coding agents: Claude Code, Codex and Cursor work from your Stride backlog over MCP.

Coding agents write plausible code for the wrong requirement when all they see is the repo and a pasted ticket. Stride runs a Model Context Protocol (MCP) server, so Claude Code, Codex and Cursor read the story with its acceptance criteria, designs and tests, then write status, comments, test results and defects back to the same record. Every call is permission-checked and audit-logged.

Why does a coding agent drift from the backlog?

Most teams hand work to a coding agent by pasting a ticket into a prompt. From then on the agent works from a snapshot: it cannot see the acceptance criteria someone edited yesterday, the design the story links to, or the tests that will judge the change. When it finishes, a person goes back to the tracker to move the ticket, summarise what changed and log what was tested, or nobody does and the backlog drifts from the code. Connecting a tracker's own MCP server removes the copy-paste but not the gap: an issue on its own still leaves out the designs, tests and decisions that live in other tools.

How do Claude Code, Codex and Cursor work from a Stride story over MCP?

Stride exposes the whole delivery record over MCP. A developer generates a project-scoped agent key in Settings → Integrations, adds the Stride server to Claude Code, Codex or Cursor, and the agent discovers the tools itself. One call, get_story_context, returns a story with its acceptance criteria, linked designs and documents, sub-tasks, test cases, defects, pull requests and recent comments. As it works, the agent moves the story across the board, comments on progress, creates test cases from each acceptance criterion, records test results and files defects, and start_story_work opens a GitHub draft pull request that links back to the story. People can hand a story, or a whole selection, to a connected agent from the backlog, and the agent pulls it with next_task.

  • One MCP endpoint for Claude Code, Codex and Cursor, connected with a project-scoped agent key
  • get_story_context: the story, acceptance criteria, linked designs, documents, tests, defects and pull requests in one call
  • Write-back as the agent works: status changes, comments, test cases from acceptance criteria, test results and defects
  • start_story_work opens a GitHub draft pull request that links back to the story
  • Hand stories to a connected agent from the backlog; the agent pulls them with next_task
  • Every call permission-checked before it runs and recorded in the audit log as via: "mcp"

Setup guides

Best for

Engineering teams already using Claude Code, Codex or Cursor who want the agent to start from the real requirement and leave the backlog up to date when it finishes.

Not for

Teams that do not use a coding agent yet, and anyone expecting agents to act unprompted: MCP is pull-based, so an agent works when someone runs it, and an agent key reaches only the projects chosen when it was created, with its creator's permissions.

Frequently asked

What is MCP project management?
Project management in which AI agents read and update the work through the Model Context Protocol (MCP), the open standard Anthropic introduced in late 2024 for connecting AI applications to external tools and data. Instead of a person pasting a ticket into a prompt, the agent fetches the story itself, works on it, and writes status, comments and results back to the same record.
Which coding agents can connect to Stride?
Claude Code, OpenAI Codex and Cursor. All three use the same Stride MCP endpoint and the same tools; only the configuration format differs: JSON for Claude Code, TOML for Codex and mcp.json for Cursor. Each has a setup guide linked on this page.
Can the agent reach every project in the workspace?
No. An agent key reaches only the projects chosen when it was created and acts with its creator's permissions, optionally capped to a lower role. Stride checks every tool call against those permissions before it runs and records it in the audit log.
Does Stride push work to the agent?
No. MCP is pull-based: a server cannot push a prompt into a running agent session. Instead you hand a story, or a whole selection, to your agent from the backlog, tell the agent to start, and it pulls the staged story with next_task, together with the full story context.
How is this different from Jira's or Linear's MCP server?
Atlassian and Linear both run official MCP servers: Atlassian's connects agents to Jira, Confluence and its other apps under your existing permissions, and Linear's lets agents find, create and update issues, projects and comments. Both give an agent the work item. Stride's difference is the record around it: acceptance criteria, test cases, designs and decisions are linked nodes, so one call returns them together, and the agent can record test results and defects against the right story.
Which plan includes the MCP server?
Every paid plan includes MCP access, the public API and webhooks. Starter includes one API or agent key; Pro and Enterprise raise that limit.
Defined in our glossary
Model Context Protocol (MCP)Agentic workflowAcceptance criteriaTraceability matrixDefinition of Done