# Autonomous Ai Agency

> Autonomous AI Agent Infrastructure Platform — OpenAI-compatible AI gateway with MCP support, multi-agent orchestration, tool calling, observability, memory, RAG, AI workflows, and unified infrastructure for any LLM provider.

- **Type:** MCP server
- **Install:** `agentstack add mcp-strikersam-autonomous-ai-agency`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [strikersam](https://agentstack.voostack.com/s/strikersam)
- **Installs:** 0
- **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [strikersam](https://github.com/strikersam)
- **Source:** https://github.com/strikersam/autonomous-ai-agency
- **Website:** https://autonomous-ai-agency.strikersam.workers.dev/

## Install

```sh
agentstack add mcp-strikersam-autonomous-ai-agency
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Autonomous AI Agency

**Your first AI hire comes with an entire agency.**

Self-hosted · Privacy-first · One URL to start

[](https://github.com/strikersam/autonomous-ai-agency/releases/tag/v5.0.0)
[](https://github.com/strikersam/autonomous-ai-agency/actions/workflows/ci.yml)
[](https://github.com/strikersam/autonomous-ai-agency/actions/workflows/deploy-backend.yml)
[](https://www.python.org/)
[](LICENSE)

**[Live Demo](https://autonomous-ai-agency.strikersam.workers.dev/) · [API Docs](https://local-llm-server.onrender.com/docs) · [Changelog](docs/changelog.md)**

---

## The problem Autonomous AI Agency solves

You're running a small or medium-sized company. You know AI could help — but the reality is messy:

- **Claude Code, Codex, and equivalent agentic platforms don't automatically do the work.** They need prompting, PR creation, test execution, CVE checks, and doc updates — and all of that requires skills setup and workflow creation that most teams never get right.
- **Your context isn't getting built.** Every prompt you send to Copilot or Claude is training data for someone else's model. The model gets smarter, but your agents don't learn your codebase, your preferences, or your business rules. Context built for your agents stays yours — it compounds over time and makes the next run faster and smarter than the last.
- **Monthly SaaS bills are stacking up.** Copilot + Notion AI + Jasper + SEO tool + monitoring tool = hundreds of dollars a month per person, and they don't talk to each other.
- **"AI agents" are mostly demos.** They hallucinate, they can't commit code, they don't remember yesterday, and they crash silently. Where AI agents genuinely help is running the full operations of a company — autonomously, end to end — in a managed platform that brings only the decisions that matter to you for approval.

Autonomous AI Agency is the answer to all of this. It is a **self-hosted AI agency** — a platform that provisions a full fleet of specialist agents for your business from a single website URL, runs them 24x7 on hardware you control, and brings only the decisions that matter to you for approval.

---

## What you get

Paste one URL. Walk away. Come back to a working AI team.

```text
https://yourcompany.com
       ↓
  [Autonomous AI Agency]
       ↓
  CEO agent + specialist fleet
  running 24x7 on your server
       ↓
  Bug fixes · PRs · blog posts · CVE scans · SEO · support replies
  — with your approval before anything ships
```

No config files. No integration wiring. No per-seat pricing. No data leaving your server.

---

## Screenshots

> Captured from a live deployment. Regenerate with `python scripts/capture_screens.py`, then `python scripts/sync_readme_gallery.py`.

_See the [full UI tour](#screens) below — every screen, desktop and mobile._

---

## Who is this for?

### The 5-person SaaS startup that can't afford a full team yet

You ship fast but quality suffers. PRs pile up. Docs go stale. Dependencies rot. The security audit you've been meaning to run has been in the backlog for six months.

With Autonomous AI Agency:
- A **Dev specialist** reviews every PR while you sleep and leaves inline comments
- A **Security specialist** runs a daily CVE scan and opens fix PRs automatically
- A **Docs specialist** keeps your README and API docs in sync with code changes — on every push
- A **Release manager** bumps the version, writes the changelog, tags, and opens the release PR — you just approve

**Result:** A team that works at night, never asks for a raise, and doesn't need you to explain the codebase.

---

### The e-commerce shop with a 10-person ops team

You're running Shopify or a custom store. Your team spends half their time on tasks a machine could do: updating product descriptions, checking for broken pages, triaging support tickets, monitoring SEO rankings.

With Autonomous AI Agency:
- An **E-commerce specialist** monitors your storefront every 30 minutes for uptime issues, TLS expiry, and stack changes
- A **PIM specialist** keeps product descriptions, attributes, and taxonomy consistent
- A **Support specialist** triages incoming tickets, drafts responses, and flags edge cases for human review
- An **SEO specialist** runs Screaming Frog-class site audits (103 checks across SEO/GEO/AEO/AIO pillars), quantifies findings as revenue-at-risk, can auto-fix issues when your repo is connected (dry-run diffs by default; `apply=true` writes changes), and delegates the rest as WSJF-prioritized tasks — see [docs/seo-audit.md](docs/seo-audit.md)
- A **Content specialist** writes product copy and blog drafts from a brief you drop in plain English

**Result:** Your ops team focuses on decisions, not repetitive maintenance.

---

### The digital agency running 10 client accounts

You manage multiple companies' tech stacks. Right now that means 10 sets of credentials, 10 monitoring setups, 10 different runbooks.

With Autonomous AI Agency:
- **Each client gets their own company profile** with its own agent fleet, knowledge graph, and schedules
- **Agents are isolated per tenant** — no cross-contamination of client data or context
- **One dashboard** to see the health of all clients, with per-client HITL approval flows
- **On-call handoff** is automatic — agents brief you when they find something, not at 3 AM when the site is down

**Result:** The same ops coverage you'd charge for a full-time hire, at marginal infrastructure cost.

---

### The professional services firm that runs on documents and tribal knowledge

Your company's knowledge lives in Slack threads, email chains, and the heads of people who might leave next month.

With Autonomous AI Agency:
- The **Knowledge specialist** reads your code, docs, Slack exports, and past decisions and builds a living internal wiki — updated automatically when things change
- The **Research specialist** monitors industry trends, competitor moves, and technology updates and drops weekly digests
- The **Portfolio manager** tracks initiatives, surfaces blockers, and runs weekly stand-up summaries
- Every agent response comes with sources and reasoning — nothing is a black box

**Result:** Knowledge that doesn't walk out the door.

---

## How it works — the 5-minute version

1. **Paste your website URL.** `https://acme-store.com`
2. **Autonomous AI Agency scans it.** Playwright + HTTP fingerprinting detects your tech stack, business systems, and integrations (Shopify, Stripe, Google Analytics, GitHub, Intercom, Salesforce, …).
3. **It asks you the right questions.** AI-generated onboarding questions based on what it found — not generic forms.
4. **Specialists are auto-provisioned.** A fleet of agents is assembled from 35 specialist families — exactly the ones your business needs, with the right runtimes and skills bound.
5. **Schedules activate.** Health scans, security audits, code quality checks, SEO monitoring, graph sync — all running on their own cadence without manual setup.
6. **You talk to the CEO in plain English.** "Fix the memory leak in issue #142." "Write the Q3 launch post." "Plan next sprint." The CEO decomposes the job, delegates to the right specialist, and returns a result with evidence — PR link, test output, diff, reasoning trace.
7. **You approve what matters.** Agents never merge code, deploy, or send external messages without your explicit sign-off. Low-risk tasks (formatting docs) can be auto-approved. High-stakes decisions (production deployments) always pause for you.

---

## The 24x7 agency — your agents never go idle

The defining feature of Autonomous AI Agency is not what agents can do — it's that they keep doing it, automatically, even when you're not watching.

### What runs automatically after onboarding

| Schedule | Cadence | What it does |
|---|---|---|
| **Website health scan** | Every 30 min | Uptime, TLS expiry, stack drift detection |
| **Security audit** | Daily 9 AM | CVE scan, security headers, repo secret scan |
| **Stack change detection** | Daily 6 AM | New frameworks, dropped libraries, new integrations |
| **Code quality scan** | Daily 12 PM | Lint, duplication, complexity, stale dependencies |
| **Trend watch** | Every 6 hrs | Model releases, framework updates, competitor tech |
| **Company graph sync** | Every 30 min | Specialist health, runtime responsiveness, schedule status |
| **Doc-sync** | On every push | API docs, architecture records, and runbooks auto-updated |

### When something goes wrong, agents fix it — not you

```text
Health scan detects broken page at 3 AM
     ↓
Security or Dev specialist creates a fix task automatically
     ↓
Agent branches, writes fix, opens PR, watches CI
     ↓
CI green + low-risk → auto-approve gate passes, PR merges
CI green + needs human eyes → surfaces to your dashboard
     ↓
You see it in the morning: "PR merged, page restored at 3:12 AM"
```

### Nothing goes down quietly

| Failure scenario | Countermeasure |
|---|---|
| Agent crashes mid-task | Crash-recovery reconciler re-queues stranded tasks on restart |
| Runtime sidecar goes to sleep | Every CEO delegation calls `RuntimeManager.wake_all_sleeping_runtimes()` first — sleeping sidecars are woken or marked `still_sleeping`, and the CEO routes around whichever stay down |
| AI session rate-limited / exhausted | `ai_runner.py` watchdog detects the gap and resumes from last checkpoint |
| LLM provider goes down | Provider chain: Bedrock → NIM → DeepSeek → Anthropic → Ollama — automatic failover |
| Missed schedule | Scheduler reconciles on boot — nothing is silently skipped |
| Context lost between sessions | Company Graph + persistent chat history give full context on every wake |

---

## Two layers, plain English: Portfolio Management vs. Loop Engineering

People often ask how the **portfolio process** differs from **loop engineering**. They sound similar — both are about the agency running itself — but they answer two *different* questions and work at two *different* layers. You need both.

> **The kitchen analogy.** Imagine a restaurant that runs itself.
> - **Portfolio Management is the manager** deciding *what to cook tonight* — which dishes are worth the kitchen's limited time, based on what sells and how much effort each takes.
> - **Loop Engineering is the kitchen line** that actually *cooks the food, tastes it, and re-cooks anything that's wrong* — on its own, every night, without someone standing over it.
>
> The manager points the kitchen at the right work; the kitchen is the machine that does the work and keeps itself running. A great manager with no kitchen ships nothing; a great kitchen with no manager cooks the wrong food.

**What they have in common**
- Both let the agency operate **without you micromanaging** every step.
- Both are **continuous** — they run on a cadence, not once.
- Both **feed each other**: the portfolio decides priorities; the loops carry them out and report back what got done.

**Where they differ**

| | **Portfolio Management** | **Loop Engineering** |
|---|---|---|
| **The question it answers** | *What should we work on, and in what order?* | *How does the work get done — and stay healthy — without me?* |
| **Layer** | Deciding / prioritising (strategy) | Doing / orchestrating (execution) |
| **How it decides** | **WSJF** — ranks big initiatives by *value ÷ effort* and lays out a Now / Next / Later roadmap | A repeating **schedule → do → check → fix → repeat** cycle with memory and self-healing |
| **Time horizon** | Weeks to quarters (epics, roadmap) | Seconds to hours (each run, continuous) |
| **If it's missing** | The agency works hard on the *wrong* things | The right work *never actually gets done or maintained* |

**The neat part:** in this repo, the portfolio process is *itself just one of the loops*. The roadmap refresh runs on its own cadence (every 6 h) like every other autonomous loop — catalogued alongside the rest in [`loops/registry.yaml`](loops/registry.yaml). So **Loop Engineering is the operating model for the whole machine, and Portfolio Management is the planning discipline that one of those loops runs to aim the others.**

One line to remember: **Portfolio Management points the machine at the highest-value work; Loop Engineering *is* the machine that runs itself.**

---

## The full agent capability roster

### Engineering

| What you say | What the agent does |
|---|---|
| "Fix the bug in issue #142" | Reads issue, reproduces, writes fix, opens PR, watches CI, awaits your approval |
| "Audit our dependencies" | Scans for CVEs, generates an upgrade plan with test coverage, opens a safe PR |
| "Review this PR" | Multi-perspective analysis: security, correctness, performance, maintainability — inline comments |
| "Write tests for the auth module" | Generates unit + integration tests with realistic fixtures |
| "Do a release" | Bumps version, writes changelog, tags, runs CI, opens release PR |
| "Refactor the payment service" | Identifies coupling issues, proposes a plan, executes on approval |
| "Keep docs in sync" | After every push: updates API docs, architecture records, runbooks automatically |

### Content & knowledge

- Write product descriptions, landing pages, blog posts, and case studies from a brief
- Summarise and classify GitHub issues, Slack threads, and support tickets
- Maintain an internal wiki — agents update pages when code or decisions change
- Weekly trend digests: new model releases, framework updates, industry moves

### Operations & DevOps

- Monitor CI/CD pipelines; alert when something needs a human decision
- Schedule daily summaries, weekly audits, and on-call handoff briefs
- Route every LLM request to the optimal local model (code → Qwen3-Coder, reasoning → DeepSeek-R1)
- Real-time health diagnostics for all agents, runtimes, and providers

### Agile, portfolio & product

- **Agentic agile**: standups, retrospectives, sprint reviews, backlog grooming — coached cadence, automated artifacts
- **Portfolio management**: roadmapping, prioritisation, resource allocation, strategy tracking
- **Delivery management**: sprint planning, release coordination, cross-team unblocking
- **Product**: turn a brief into user stories, acceptance criteria, and a prioritised backlog

### Business & domain specialists (auto-provisioned from the URL scan)

| Detected system | Specialist provisioned |
|---|---|
| Storefront / commerce stack | **E-commerce** · **Merchandising** · **OMS** |
| Product catalog / PIM | **PIM** (product data, attributes, taxonomy) |
| Media / asset platform | **DAM** (ingestion, metadata, delivery) |
| CRM / support desk | **CRM operations** · **Support** (triage, KB, SLA) |
| Analytics / search / SEO | **Analytics** · **SEO** · **Content strategist** |
| Marketing automation | **Marketing** (campaigns, attribution, A/B) |
| Research / market data | **Trading & market research** · **Research** |
| Cloud / infra / CI | **Platform operations** · **DevOps** · **Security** · **CI/fix** |

> **35 specialist families total** — engineering + business + domain. Each family has typed I/O contracts, an optimal runtime, and bound Skills.

---

## The skill library — superpowers agents can call

Every specialist can call typed, versioned Skills on demand:

| Skill | What it does |
|---|---|
| **ECC** | Orchestrate other AI harnesses: Claude Code, Cursor, Codex, OpenCode, Aider |
| **Graphify** | Query your entire codebase as a knowledge graph — 70x fewer tokens than reading files |
| **Council Review** | Multi-perspective diff review: security / correctness / performance / maintainability — structured APPROVE / REJECT verdict |
| **Obsidian Knowledge Graph** | BFS traversal, connected components, tag search over your internal wiki |
| **Dependency Audit** | CVE scan + safe upgrade PR generation |
| **Agentic Agile** | Sprint ceremonies, retros, standups as a coached cadence |
| **Financial Analyst** | Burn rate, runway, gross margin, ROI-based budget reallocation |
| **Release Readiness** | Gate check before an

…

## Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [strikersam](https://github.com/strikersam)
- **Source:** [strikersam/autonomous-ai-agency](https://github.com/strikersam/autonomous-ai-agency)
- **License:** MIT
- **Homepage:** https://autonomous-ai-agency.strikersam.workers.dev/

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** yes
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** yes
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/mcp-strikersam-autonomous-ai-agency
- Seller: https://agentstack.voostack.com/s/strikersam
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.