> ## Documentation Index
> Fetch the complete documentation index at: https://docs.pnbr.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Quickstart

> Connect the Runtime MCP, save a source, and ask your agent to retrieve it again.

Your first loop does not require code or a shape. You will connect an agent, save
a short source in the active project, and retrieve its exact wording.

## Before you start

[Connect the Runtime MCP](/getting-started/connect-agent) as `penumbra`, then
complete the browser sign-in and confirm that the server is connected.

## Save and retrieve a source

<Steps>
  <Step title="Check the active project">
    Tell your agent:

    > Use Penumbra. Tell me which project is active and what is already in it. Do not make any changes.

    Confirm that the project is where you want the source to live.
  </Step>

  <Step title="Save the source">
    Give the agent a small piece of useful material:

    > Save this as a source titled “Procurement note”: Enterprise buyers stall on procurement, not price. Tell me when it is ready and give me its source id. Do not extract it yet.
  </Step>

  <Step title="Read it back">
    Ask:

    > Read the Procurement note and quote what it says about enterprise buyers.

    The answer should contain the wording you saved, not a newly generated summary.
  </Step>
</Steps>

<Check>
  You have completed the basic Runtime loop: orient, add evidence, and retrieve it again.
</Check>

<Accordion title="What the agent did">
  ```text theme={null}
  penumbra_status
  → penumbra_add_source
  → penumbra_source_rag
  ```
</Accordion>

## If it did not work

* **Wrong project:** ask the agent to show the available projects and switch before saving again.
* **Source is still processing:** the agent should poll the receipt returned by `penumbra_add_source` rather than submit the source again.
* **You received a summary instead of a quote:** ask the agent to use verbatim source retrieval.

## Add structured knowledge next

Sources preserve evidence. Shapes define the structured knowledge your agents can
extract or capture from that evidence.

## Where to go next

<Columns cols={2}>
  <Card title="Extract a source" icon="wand-magic-sparkles" href="/runtime/extract-source">
    Turn saved evidence into structured knowledge through a shape.
  </Card>

  <Card title="Capture knowledge" icon="pen-to-square" href="/runtime/capture-knowledge">
    Stage something you already know and validate it before submission.
  </Card>

  <Card title="Create a shape" icon="shapes" href="/shapes/create-shape">
    Describe the concepts and relationships that matter in your domain.
  </Card>
</Columns>