You can go from a blank slate to an active research graph in a few minutes. This page walks you through the core Thesis flow: signing in, creating a node, indexing a source, talking to the agent, and, if you want agent-host integration, running the one-command MCP setup.
This is the Thesis quickstart (the research workspace). For the Synsci CLI terminal coding agent, see the Synsci CLI quickstart.
1

Sign in to Thesis

Go to thesis.syntheticsciences.ai and sign in. Thesis provisions your account automatically on first login, there is no separate registration step.After signing in you land on the Dashboard, which shows your research graphs and a spotlight search bar.
2

Create a project node

From the Dashboard, click the New graph button (or the floating action button on the canvas). Give your root node a title that describes the research topic or question you are starting from.Every node in Thesis is either staged (editable, work in progress) or committed (frozen as durable research history). Start staged, you can commit once findings are stable.
Think of a root node as the top of a research tree. You will branch child nodes from it as experiments, source summaries, and insights accumulate below it.
3

Add a knowledge source

Open the Sources panel from the sidebar (or navigate to /sources). Paste the URL of a paper, repository, documentation site, or dataset you want Thesis to index.Once indexed, the source becomes searchable by the agent, Oracle jobs, and unified search. You can also attach source chunks directly to nodes as evidence.
Full external source indexing requires the knowledge layer to be active. If you are on a plan that includes knowledge features, indexing starts immediately after you add a URL.
4

Open the agent and chat

Navigate to the Agent page (/agent) from the sidebar. The agent panel has three tabs:
  • Canvas, the graph canvas showing your nodes, edges, and staged work.
  • Agents, a live view of running or recent sub-agent instances and their telemetry.
  • Files, the project filesystem for artifacts, reports, and sub-agent outputs.
Type a message to the agent. Unlike a generic chatbot, the Thesis agent hydrates from your actual graph state before responding. It reads your nodes, research log entries, project files, and indexed sources. Ask it to summarize what you have so far, suggest next experiments, or search for relevant prior work.
You can open the Research modal from anywhere in the app with Cmd+Shift+K (or Ctrl+Shift+K on Windows/Linux). It gives you quick access to Ask, Oracle, Tracer, and Document search without leaving your current view.
5

Stage your first experiment (optional)

Ask the agent to design an experiment for your root hypothesis. The agent will propose an 8-section blueprint:
  1. Hypothesis restatement
  2. Falsification criterion
  3. Method
  4. Expected outcomes
  5. Baseline and control
  6. Metrics
  7. Data configuration
  8. Rigor review
The proposal appears as a card in the chat. Approve it to mint the token required to run compute; reject it to log the decision as a dead end and iterate.
Approving an experiment proposal authorizes the agent to spawn a compute-spending sub-agent. Review the blueprint carefully before approving, the approval is explicit and logged.
6

Connect an MCP host (optional)

If you want your AI coding agent. Claude Code, Codex, Cursor, OpenCode, Synsci CLI, or another supported host, to read and write your Thesis graph directly, run the @synsci/thesis MCP helper:
npx --yes @synsci/thesis setup
The CLI will:
  1. Detect which supported hosts are installed on your machine.
  2. Open your browser to authenticate and provision an API key.
  3. Write the Thesis MCP server configuration into each detected host’s config file.
After setup, your agent host has access to all 86 Thesis MCP tools: node lifecycle, graph topology, knowledge search, artifact management, compute leases, export, and more.
npx --yes @synsci/thesis setup
On a headless or SSH machine, use device auth instead:
npx --yes @synsci/thesis setup --auth-mode device
For web-based MCP hosts (Claude.ai, ChatGPT.com), paste this URL when prompted for a connector:
https://thesis.syntheticsciences.ai/mcp-server

​
What’s next