Most AI tools are strangers. Mission Squad is the agent that knows your stack — your Asana boards, your Drive folders, your Gmail, your Notion, your private data — and does the work you'd otherwise be doing at midnight.
Become the operator your team can't replace.
The work that buried your Tuesday now runs on a schedule, with a full trace, while you do the work only you can do. This isn't a chatbot. It's a way of operating.
What you start doing
What you stop tolerating
Describe the squad. It builds it.
You don't assemble agents by hand. You tell the configuration agent what you want, and it builds the whole thing for you.
Ask for "a factory that searches YouTube for a topic, pulls transcripts for the top five videos, and writes one holistic summary," and it doesn't just start typing. It checks what you already have, then proposes a plan: each agent, the model it runs on, the tools it needs, and how they hand off to each other. You approve the plan. It writes the prompts, creates the agents, and wires up the factory.
Changed your mind? Tell it — "use GPT 5.4 mini instead of Sonnet" — and it reconfigures in place. No rebuild. No config files. No clicking through twelve forms. Get started in the docs →
msq_list_servers{}msq_list_server_tools{"serverName":"youtube-transcripts"}I have analyzed your request and designed a sequential multi-step Factory to accomplish this. Here is the proposed plan:
1. Proposed Agents
Agent 1: YouTube Searcher (youtube-searcher)
- Description: Searches YouTube for phrases similar to the user's input, performs multiple searches to find the best results, and returns video titles and links.
- Model:
claude-sonnet-4-6:1m - Tools:
search_youtube(from theyoutube-transcriptsMCP server)
Agent 2: YouTube Transcript Summarizer (youtube-transcript-summarizer)
- Description: Receives a list of YouTube video links, fetches transcripts for the 5 most relevant videos, summarizes each, and combines them into a holistic summary.
- Model:
claude-sonnet-4-6:1m - Tools:
get_youtube_transcript(from theyoutube-transcriptsMCP server)
2. Proposed Factory Structure
- Factory Name: YouTube Topic Research Factory
- Description: Sequential research pipeline that searches YouTube for a topic, retrieves transcripts for the top 5 videos, and synthesizes a holistic summary.
- Steps:
youtube-searcher→youtube-transcript-summarizer
Please let me know if you approve of this plan, and I will proceed with generating the prompts, creating the agents, and configuring the factory!
msq_generate_prompt{"type":"agent","name":"youtube-searcher","model":"claude-sonnet-4-6:1m"…}msq_generate_prompt{"type":"agent","name":"youtube-transcript-summarizer","model":"claude-sonnet-4-6:1m"…}msq_add_agent{"name":"youtube-searcher","model":"claude-sonnet-4-6:1m","systemPromptId":"…","tools":["search_youtube"]}msq_add_agent{"name":"youtube-transcript-summarizer","model":"claude-sonnet-4-6:1m","systemPromptId":"…","tools":["get_youtube_transcript"]}msq_create_factory{"name":"YouTube Topic Research Factory","steps":[{"kind":"agent","agentRef":"agent/…"},{"kind":"agent","agentRef":"agent/…"}]}I have successfully created and configured your YouTube Topic Research Factory. Here is what was set up in your MissionSquad account:
- Agent created —
youtube-searcher - Agent created —
youtube-transcript-summarizer - Factory created — YouTube Topic Research Factory
Real config session — request → proposed plan → created agents + factory
It reads your stack first.
Before it proposes anything, it lists your connected MCP servers and the tools they expose — so the plan fits what you have.
It proposes before it builds.
You see the agents, models, and hand-offs as a plan you approve — not a black box that silently spins up infrastructure.
It edits by conversation.
Swap a model, rewire a step, rename an agent. The squad changes as fast as you can describe the change.
It already speaks to the tools you live in.
Connect a server, authorize, and your squad can read and act in the systems you already use — Google Workspace, Gmail, Notion, Asana, Dropbox, Shopify, and dozens more. Need something we don't have yet? Drop in your own OAuth compatible MCP server and it shows up alongside the rest. Read the MCP guide →
All connections · OAuth 2.1 · DCR · PKCE · audited permission scopes
Bring your own brains. Run it on your own metal.
Mission Squad isn't married to one model vendor. Point an agent at OpenAI, Anthropic, Google, Cohere, Grok, Mistral — or run open models on hardware you own. Swap the model behind any agent in seconds, with no rewrite and no SDK to rip out.
And it runs where you do. Deploy in your own VPC or on-prem so your keys, your data, and your traffic never leave your control. "Your data" means your data — not a polite word for someone's training set. See the self-hosting guide →
Fan out, or chain it.
Mission Squad gives you two ways to compose agents. A workflow fans out across agents working in parallel; a factory chains steps in sequence. Many systems use both. Workflows → · Factories →
Several agents take on the same request in parallel. Each works independently, then every response is fed back into one main agent that reads them all and produces a single answer.
Steps run in sequence. Each step's output becomes the next step's input — and any step can be a single agent or an entire workflow nested inside.
Workflows fan out · factories chain · compose them however the work demands.
Watch the work, not a status bar.
Most platforms show you a polished demo. We'll show you a real run — every agent that fired, every tool it called, the exact arguments it passed, and the data that came back.
Here's a financial-analysis workflow: a single request fans out to nine specialist agents — fundamentals, sentiment, macro, and six technical analysts — each pulling its own data and writing its own section, before a lead agent stitches them into one report. Open any sub-agent and you can read the tool calls underneath: fundamentalData_companyOverview, the income statement, the balance sheet, the cash flow — and the JSON each one returned.
Every step is counted. When a run does something you didn't expect, you don't guess — you open the trace and read exactly what happened.
fundamentalData_companyOverview{"symbol":"AAPL"}fundamentalData_incomeStatement{"symbol":"AAPL"}fundamentalData_balanceSheet{"symbol":"AAPL"}fundamentalData_cashFlow{"symbol":"AAPL"}{
"toolName": "fundamentalData_earnings",
"serverName": "alphavantage",
"args": { "symbol": "AAPL" }
}Apple is a large-cap, highly profitable integrated hardware/software/services company with a dominant premium device position and expanding recurring-services revenue. Fiscal 2025 results show revenue of $416.2B and net income of $112.0B…
Apple Inc. (AAPL) presents as a high-quality, market-leading technology firm demonstrating exceptional financial performance and strong short-term market momentum…
Model-generated analysis, not investment advice.
Plate #1 · Short-term financial analysis — 9 sub-agents · full tool-call trace · per-step token accounting · kept in run history alongside every prior run.
Your agents will run when you need them. And when they don't, you will know why.
Pick the one that fits where you are today.
Get the field briefing.
A weekly note from the Mission Squad lab: workflows we built, what broke, what we learned, what we'd do differently. No pitch. No sales follow-up. Unsubscribe in one click.
Deploy your first squad.
Free to start. No credit card. Connect one tool, describe one agent, and read the trace yourself. If you're not up and running in an afternoon, we got it wrong and want to hear about it.