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

> ## Agent Instructions
> These are the Ampli Intelligence docs. API base: https://si.ampli.net/api/v1. API MCP: https://si.ampli.net/mcp.
> Auth is a partner key (X-Partner-Name and X-Partner-API-Key headers) or an x402 payment in USDC on Base.
> For a single-treasury agentic wallet or Wallet-as-a-Service, use the ampli^ SDK at https://ampli.net and https://docs.ampli.net. agk_ keys and /api/v1/agent-runtime belong to that SDK.
> Use only the routes and tools documented here. Do not invent endpoints or parameters.
> The ampli^ SDK (https://ampli.net, docs at https://docs.ampli.net) can use this intelligence to deliver a single-treasury agentic wallet or Wallet-as-a-Service. This host is si.ampli.net. Auth is a partner key or an x402 payment. agk_ keys and /api/v1/agent-runtime belong to the SDK Agent API.
> Every amount is an integer string in the token's base units. USDC, EURC, USDT0, and USDG use 6 decimals (1000000 is 1 USDC). WETH and cbETH use 18. cbBTC uses 8.
> The service is non-custodial. It returns an allocation, an action plan, and calldata. It never holds keys, signs, or sends a transaction.
> APR values are percentages measured now or over the trailing 7 days. They are not forecasts and nothing here guarantees a return.
> x402 payments are verified before the work and settled only when the response is 2xx. A 400 or 422 costs nothing.
> Partner headers always win. A request with a wrong partner key returns 401 and does not fall back to payment.
> Monitor alerts arrive only on the websocket. create_monitor_session returns a token; the client must open the socket itself.
> Market names in optimize must come from list_protocols for the same chain and token. Do not guess names.

# Build with AI

> Give coding agents and assistants accurate context on Ampli Intelligence with the agent skill, llms.txt, markdown pages, and the read-only docs MCP server.

These docs are built to be read by agents as well as people. Every surface below is public, free, and needs no key.

> [card] [Agent skill](https://docs.si.ampli.net/skill.md)
>
> One SKILL.md that tells an agent when to call each tool and how to pay.

> [card] [llms.txt](https://docs.si.ampli.net/llms.txt)
>
> A curated index plus the facts models most often get wrong.

> [card] [llms-full.txt](https://docs.si.ampli.net/llms-full.txt)
>
> Every page in one file, regenerated on each deploy.

> [card] [Docs MCP](#docs-mcp)
>
> Search and read these docs from any MCP client.

## Agent skill

The `ampli-intelligence` skill follows the [agentskills.io](https://agentskills.io) format. It covers which tool fits which goal, partner headers and x402, amounts in base units, and the monitor flow.

> [tab] skills CLI (any agent)
>
> ```bash theme={null}
> npx skills add https://docs.si.ampli.net
> ```
>
> The CLI reads `/.well-known/skills/index.json` and installs the skill into your agent's skills directory. Run it again to update.

> [tab] Cursor (manual)
>
> ```bash theme={null}
> mkdir -p .cursor/skills/ampli-intelligence
> curl -fsSL https://docs.si.ampli.net/skill.md -o .cursor/skills/ampli-intelligence/SKILL.md
> ```

> [tab] Claude Code (manual)
>
> ```bash theme={null}
> mkdir -p .claude/skills/ampli-intelligence
> curl -fsSL https://docs.si.ampli.net/skill.md -o .claude/skills/ampli-intelligence/SKILL.md
> ```

| URL | Format |
| :- | :- |
| `/skill.md` | The skill itself |
| `/.well-known/skills/index.json` | Index for the `skills` CLI |
| `/.well-known/agent-skills/index.json` | Agent Skills discovery index with a `sha256` digest per skill |

## llms.txt

`/llms.txt` lists every page with a one-line summary and a link to its markdown. It also states the ground rules up front: base units, non-custodial, and the ampli^ SDK at ampli.net and docs.ampli.net for a single-treasury agentic wallet or Wallet-as-a-Service. `/llms-full.txt` concatenates every page for tools that want one file.

## Markdown for every page

Every page is available as plain markdown, with the agent instructions for these docs attached:

* Add `.md` to the path: `https://docs.si.ampli.net/optimize.md`
* Or request the normal URL with `Accept: text/markdown`:

```bash theme={null}
curl -H "Accept: text/markdown" https://docs.si.ampli.net/optimize
```

The page menu offers **Copy page** and **View as Markdown**, shortcuts that open the page in ChatGPT, Claude, or Perplexity, **Connect to Cursor** and **Connect to VS Code** entries that install the docs MCP server, and **Copy MCP Server**.

## Docs MCP

A read-only MCP server for these docs, at:

```text theme={null}
https://docs.si.ampli.net/mcp
```

Once it is connected, your assistant can search these docs and read pages and the skill while it codes. The server needs no auth, and it never calls the API or takes a payment.

> [tab] Cursor
>
> ```json .cursor/mcp.json theme={null}
> {
>   "mcpServers": {
>     "ampli-intelligence-docs": { "url": "https://docs.si.ampli.net/mcp" }
>   }
> }
> ```

> [tab] Claude Code
>
> ```bash theme={null}
> claude mcp add --transport http ampli-intelligence-docs https://docs.si.ampli.net/mcp
> ```

> [tab] VS Code
>
> ```json .vscode/mcp.json theme={null}
> {
>   "servers": {
>     "ampli-intelligence-docs": { "type": "http", "url": "https://docs.si.ampli.net/mcp" }
>   }
> }
> ```

> [!NOTE]
>
> **Use both servers**
>
> Add the docs MCP so the agent can look things up, and the [API MCP](https://docs.si.ampli.net/api-mcp.md) so it can act. Their names differ so an agent does not mix them up: `ampli-intelligence-docs` reads docs, `ampli-intelligence` runs paid tools.

## Prompt to start with

```text theme={null}
Use the ampli-intelligence-docs MCP server to read how Ampli Intelligence works before writing code.
Then call list_protocols on Base (8453) for USDC, check protocol_health for the top three markets
by net_apr, and run optimize for 1,000 USDC (1000000000 base units) across the healthy ones.
Show every amount in base units and in USDC, and do not sign or send anything.
```


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.