# Amp docs: AI Agent Readiness Score 68.8% (D+)

**68.8% · 55/80 · AI Agent Readiness Score · 25/30 reading points · 30/50 agent surface points**

Amp received 10 PASS votes and passed 3 of five agent surface checks. The clearest finding came from the find the exact limits task.

- Tested: 2026-08-20
- Published: 2026-09-21
- Battery: v1
- Scoring: reading 30 pts · surface 50 pts
- Docs: https://ampcode.com/manual

Three AI models, GPT 5.6 Sol, Claude Opus 5, and DeepSeek v4 Flash, each read Amp’s public documentation independently and attempted five first-hour developer jobs: run the first Amp thread, find the exact limits, recover a timed-out SDK execution, authenticate a remote MCP server, use the Amp SDK.

No accounts, API calls, or code execution were used. Every verdict came from public pages and every published quotation passed a live verification check. Preflight substitutions: T3: SDK execution errors and timeout handling replace HTTP 429. T4: Remote MCP OAuth authentication replaces webhooks.

## Freshness

[How rechecks work](https://docsforagents.com/methodology/#freshness)

- Category: [Coding agents](https://docsforagents.com/grades/?category=coding-agents)
- Tested: 2026-08-20
- Quotes verified: 2026-08-20
- Surface rechecked: 2026-09-21

No change since the test.

2 of 5 quoted passages still appear on the live pages. The quoted page changed for Run the first Amp thread, Find the exact limits, and Authenticate a remote MCP server.

## Agent surface checks · 30/50

| Check | Verdict | Points |
| --- | --- | --- |
| llms.txt | PASS | 10 |
| llms-full.txt | FAIL | 0 |
| Markdown mirror | PASS | 10 |
| MCP server | PASS | 10 |
| Docs AI | FAIL | 0 |

## The Reading Test · 25/30

| Task | GPT 5.6 Sol | Opus 5 | DeepSeek v4F | Consensus |
| --- | --- | --- | --- | --- |
| Run the first Amp thread | PASS | PARTIAL | PASS | PASS |
| Find the exact limits | PARTIAL | PASS | PARTIAL | PARTIAL |
| Recover a timed-out SDK execution | PASS | PARTIAL | PARTIAL | PARTIAL |
| Authenticate a remote MCP server | PASS | PASS | PASS | PASS |
| Use the Amp SDK | PASS | PASS | PASS | PASS |

Docs platform: Custom SvelteKit (unscored) · verified 2026-08-20

## What to fix first

These 4 fixes could add up to 24 points to the AI Agent Readiness Score. The list ranks each fix by the points it would add. [How the ranking works](https://docsforagents.com/methodology/#what-to-fix-first)

1. **+10 points · llms-full.txt · check failed**

   **Found:** The URL redirects to sign-in and returns no public corpus.

   **Fix:** Publish llms-full.txt at the docs root with the full text of every docs page in one plain-text file.

   **Evidence:** [ampcode.com/llms-full.txt](https://ampcode.com/llms-full.txt)

2. **+10 points · Docs AI · check failed**

   **Found:** The live manual exposes no public documentation assistant.

   **Fix:** Add an assistant to the docs site that answers questions from the docs and links to its sources.

3. **+2 points · Find the exact limits · PARTIAL**

   **Found:** The pricing page calls included usage a minimum and gives no rule for extra usage. DeepSeek also found orb-hour rates only in the orbs manual.

   **Fix:** Document how extra included usage is decided, and put the orb-hour overage rate on the pricing page.

   **Evidence:** [ampcode.com/pricing](https://ampcode.com/pricing)

4. **+2 points · Recover a timed-out SDK execution · PARTIAL**

   **Found:** Opus 5 found the SDK reference does not say what execute() does on abort. DeepSeek found the timeout example shows no recovery step.

   **Fix:** Document what execute() does when the signal aborts, and add a retry step to the Cancellation and Timeouts example.

   **Evidence:** [ampcode.com/manual/sdk](https://ampcode.com/manual/sdk)

## What the docs get right

- **Authenticate a remote MCP server: 3 PASS votes.** The section gives fallback client settings, add and login commands, browser authentication, token storage, and logout recovery.
- **Use the Amp SDK: 3 PASS votes.** Quick Start gives npm install @ampcode/sdk, the token export, a minimal execute() loop, and a result extraction helper, and the SDK's mode names and system/assistant/result message types match the CLI's agent modes and --stream-json output, though the SDK's session_id is presented as an execution identifier and only the manual Appendix reveals it is the Amp thread ID.
- **3 of 5 agent surface checks.** Present: llms.txt, markdown mirrors, an MCP server. Missing: llms-full.txt, docs AI.

## Run the first Amp thread

**PASS**

PASS consensus from 2 PASS, 1 PARTIAL.

The manual gives sign-in, platform installers, CLI launch, and one-thread-per-task prompting examples in one documented sequence. The Get Started list, the install commands, and the How to Prompt guidance are all on the manual page, but the manual never documents how the CLI itself authenticates: the interactive command amp login appears only on the SDK page, the manual offers only AMP_API_KEY for non-interactive use, and the linked /install page returns a sign-in wall rather than instructions. The Get Started section gives a three-step numbered list from sign-in through running the CLI, and the CLI section shows that run amp alone starts interactive mode.

## Find the exact limits

**PARTIAL**

PARTIAL consensus from 1 PASS, 2 PARTIAL.

The pages define every requested rule, but possible extra subscription usage has no formula or exact ceiling. All six values are stated with scope across two linked pages: the manual gives zero markup for individuals and non-enterprise workspaces, 50% for Enterprise, twelve-month credit expiry, no rollover, and pooled workspace credits, while /pricing supplies the per-tier included amounts ($20 on Megawatt, $200 on Gigawatt), which its own FAQ then reframes as a floor rather than a fixed figure. Included usage, credit expiry, rollover, and pooling are explicit, but the overage rate for orb hours beyond subscription is not stated as a per-unit price on the pricing page, it must be inferred from the orbs manual's per-hour rates, and Megawatt's 750 hours of orbs does not specify which orb size the limit applies to.

## Recover a timed-out SDK execution

**PARTIAL**

PARTIAL consensus from 1 PASS, 2 PARTIAL.

The example passes AbortSignal.timeout into execute, converts aborted runs to timeout errors, and rethrows every other error. The Cancellation and Timeouts example shows AbortSignal.timeout plus a signal.aborted check, and the TypeScript reference defines ErrorResultMessage, but the reference describes signal only as "Signal for cancellation support" and never states what execute() does when aborted, so the example's catch block only works if the async iterable throws on abort, which is nowhere documented, and no error subtype represents a timeout. The SDK documents error detection via is_error, timeout via AbortSignal.timeout, and distinguishing the two via signal.aborted, but the example throws on both conditions instead of showing recovery patterns such as retry, state resume, or reconnection.

## Authenticate a remote MCP server

**PASS**

PASS consensus from 3 PASS.

The section gives fallback client settings, add and login commands, browser authentication, token storage, and logout recovery. The MCP section covers the whole path in order: automatic client registration for servers like Linear, then manual registration with the exact redirect URI http://localhost:8976/oauth/callback, amp mcp add, a full amp mcp oauth login invocation with client id, secret, and scopes, token storage, and amp mcp oauth logout to recover from a stale or revoked token. The manual gives step-by-step CLI commands for manual OAuth registration, the redirect URI and scopes to configure, and how to recover from stale tokens with amp mcp oauth logout.

## Use the Amp SDK

**PASS**

PASS consensus from 3 PASS.

The quick start installs @ampcode/sdk, calls execute, iterates system, assistant, and result messages, then reads the final result. Quick Start gives npm install @ampcode/sdk, the token export, a minimal execute() loop, and a result extraction helper, and the SDK's mode names and system/assistant/result message types match the CLI's agent modes and --stream-json output, though the SDK's session_id is presented as an execution identifier and only the manual Appendix reveals it is the Amp thread ID. The SDK Quick Start documents installation, API key setup, a minimal execute() call, message streaming over three typed channels, and result extraction, all matching the CLI's execute mode, mode names, and configuration concepts.

## The receipt

> We give at least the subscription cost per month in agent usage. Depending on your usage patterns, you may get extra usage.

The pages define every requested rule, but possible extra subscription usage has no formula or exact ceiling.

- [ampcode.com/pricing](https://ampcode.com/pricing)

## Agent surface notes

Initialize returned an MCP Bearer OAuth challenge with valid resource metadata.

The live manual exposes no public documentation assistant.

## Method note

This is a reading test of public documentation, not an execution test. No accounts were created and no API calls were run. The AI Agent Readiness Score counts fifteen reading votes at PASS 2, PARTIAL 1, and FAIL 0, for 30 possible points. Five agent surface checks add 10 points each. The total is 80. Consensus chips show each row majority and do not affect scoring. The panel split on 3 of five tasks. Quotes shown here were re-fetched and confirmed verbatim on 2026-08-20.

Methodology: https://docsforagents.com/methodology/

Canonical URL: https://docsforagents.com/reports/amp-docs-ai-agent-readiness/