Make your AI agent follow instructions

Stop repeating yourself. Describe the task β€” your agent picks or builds the right workflow and follows it step by step. Works with Claude Code, Cursor, ChatGPT, and any MCP client.

Sound familiar?

  • "Write the code" β†’ the agent forgets the tests
  • "Research the topic" β†’ it invents facts and cites nothing
  • "Write a PRD" β†’ the edge cases are missing
  • "Write the article" β†’ the structure falls apart halfway
  • "Generate tests" β†’ only the happy path is covered

So you watch every step: remind, check, repeat.

Moira takes that job off your hands.

You don't need to learn to build workflows

Moira is agent-first: you describe the task, your agent runs the process.

  1. 1 Connect your agent Add Moira once to Claude, ChatGPT, Cursor or any other MCP client β€” it is one server address in the settings. Your tools stay as they are.
  2. 2 Describe the task Say what you need in plain words, the way you would tell a colleague.
  3. 3 Your agent picks the workflow It runs a ready workflow that fits, or builds one for your task through the Workflow Management Flow β€” simple or complex, whatever the task needs.

Need a workflow of your own? You don't write it by hand β€” ask your agent in plain words, for example: "Make me a workflow for reviewing a pull request: security first, then the tests."

Diagrams, variables and editing by hand are optional β€” they are there when you want to look inside.

What Moira guarantees

  • One step at a time

    The agent sees only the current step and finishes it before the next one. It can't skip ahead.

  • Checked before moving on

    Every step says what the agent must hand back. Moira checks that the answer has all of it and sends it back for a fix until it does.

  • Nothing forgotten

    Tests, docs, commits: the workflow keeps the list even when the agent loses track.

  • The same result every time

    One workflow, one standard of work. No more "it worked yesterday".

What a workflow looks like

Three tiny learning workflows. Each step is the exact instruction your agent receives; Moira checks the answer before the next one.

Example 1: Simple Steps

A straight line: one step at a time, no choices.

  1. 1 Step 1

    Read the user's request and restate the task in one sentence. If something important is unclear, ask the user before you go on.

  2. 2 Step 2

    Do the task. Keep to what the user asked for and do not add extra work.

  3. 3 Step 3

    Compare the result with the request. Fix anything that does not match before you go on.

  4. 4 Step 4

    Tell the user in a few sentences what you did and how you checked it.

moira/example-simple-steps

Example 2: One Choice

One fork: a yes-or-no answer decides what happens next.

  1. 1 Step 1

    Do the task the user asked for.

  2. 2 Step 2

    Check the result against the request. Answer yes if it fully matches, or no if something is missing.

    yes β€” it matches
    Step 3 Tell the user in a few sentences what you did.
    no β€” something is missing
    Step 4 Tell the user what is missing and why, and suggest what to do next. Do not start the task again in this flow.

moira/example-one-choice

Example 3: Several Paths

Several paths: one answer picks one of three short routes.

  1. 1 Step 1

    Read the user's request and decide what it is: a question, a bug to fix, or a new idea.

    One of
    a question
    Step 2 Answer the question briefly and say where the answer comes from.
    a bug
    Step 3 Reproduce the bug, fix it, and show the user that it no longer happens.
    an idea
    Step 4 Write a short plan for the idea, at most five steps, and ask the user whether to go ahead.

moira/example-several-paths

Try one β€” say to your agent: "Use Moira to run Example 1: Simple Steps for this task: …"

Walk through all three in the tutorial β†’

Ready workflows for everyday tasks

Universal workflows for clear tasks with no special logic. Usually your agent picks one for you.

Quick Task

Plans the work, does it with evidence and gets an independent review before handing over the result.

Pick it when a clear, bounded task that is not a code change: a document, research, a set of edits you want checked.

Robust Task

The same idea for long or critical work: retries, recovery after interruptions, replanning when facts change, honest partial results.

Pick it when the task is large or important, or likely to run into problems.

Todo List

Works through a checklist one item at a time β€” the one you give it, or one it writes once from your goal β€” with a short proof for each.

Pick it when the steps are clear and just need doing in order.

Software Development Flow

The full change to a code repository: plan β†’ code β†’ tests β†’ docs β†’ review.

Pick it when the task is a feature, a fix or a refactor in code.

Just say: "Use Moira Quick Task for this: …"

Workflows for your role

The catalog also has workflows for specific jobs. A few of them:

Architecture Design Flow

Before the code: requirements, architecture decisions and their documentation, checked by an independent review.

Say: "Use Moira Architecture Design Flow for this service: …"

A workflow of your own

Three ways to get one β€” the first is usually enough.

  1. 1

    Describe it in words

    Tell your agent what the workflow should do. It builds it through the Workflow Management Flow.

    "Make me a workflow for the weekly report: collect the numbers, write the summary, compare with last week"

  2. 2

    Start from a ready one

    Take a workflow that is close and ask the agent to adapt it.

    "Take Quick Task and add a security check before the review"

  3. 3

    Adjust it by hand (optional)

    Open any workflow in the web app and change a step yourself β€” only if you want to.

    Open the workflow and reword one step's instruction

Example conversation

You: I want a release workflow: changelog, version bump, tag, then an announcement.

Agent: Done β€” I built "Release" with four checked steps: changelog, version, tag, announcement. Shall I run it for the next release?

Built with Moira

Moira is developed with its own Software Development Flow β€” hundreds of runs on real production code.

PLAN CODE TEST DOCS GIT NOTIFY

Every step checked. Nothing skipped.

See the Software Development Flow β†’

Stay informed in Telegram

Moira messages you when something needs your attention.

Decision needed

The workflow is waiting for your answer

Step completed

A milestone is reached β€” see the result

Something went wrong

An error that needs a look

Example notification

M

MCP Moira Bot

bot

βœ… Software Development Flow

Step 4/6 completed: tests passed (42/42)

Just now

No need to keep checking β€” Moira tells you when it's done.

Run it yourself β€” Moira is open source

Self-host the engine, the web app and the MCP server in one Docker container under Apache-2.0. Free, single-tenant, and your data stays with you. Rather not run it? Use Moira Cloud.

Frequently asked questions

Quick start

Connect MCP Moira in under 2 minutes

Recommended: Use CLI command

Terminal
claude mcp add --transport http moira https://moira-mcp.com/mcp

Then authenticate:

OAuth Flow
# After adding, authenticate within claude
/mcp
# β†’ Select "moira"
# β†’ Click "Authenticate"
# β†’ Browser opens for OAuth
Alternative: Manual JSON config
# Alternative: Manual JSON config
# ~/.config/claude/mcp.json
{
  "mcpServers": {
    "moira": {
      "type": "http",
      "url": "https://moira-mcp.com/mcp"
    }
  }
}

# Then: /mcp β†’ Authenticate
Authentication without OAuth

For CI/CD, Docker, or environments without a browser β€” use an API token instead of OAuth.

1. Log in to Moira web UI β†’ Settings β†’ API Tokens
2. Create a token (starts with moira_)
3. Replace moira_YOUR_TOKEN below with your token
~/.config/claude/mcp.json
{
  "mcpServers": {
    "moira": {
      "url": "https://moira-mcp.com/mcp",
      "headers": {
        "Authorization": "Bearer moira_YOUR_TOKEN"
      }
    }
  }
}

What happens next?

  1. 1

    Type in your AI chat: "Start user onboarding flow"

  2. 2

    Your agent walks you through a short interactive onboarding that shows Moira on a real workflow.

  3. 3

    You see the available workflows and start your first real task.

Ready to try it?

Connect your agent and describe your first task.

Beta version

MCP Moira is currently in active development. Features may change without notice, data loss is possible during updates. Some workflows may behave unexpectedly. Not recommended for production-critical tasks.