# Roboflow docs: AI Agent Readiness Score 76.2% (C)

**76.2% · 61/80 · AI Agent Readiness Score · 21/30 reading points · 40/50 agent surface points**

Roboflow received 8 PASS votes and passed 4 of five agent surface checks. The clearest finding came from the find the exact limits task.

- Tested: 2026-08-19
- Published: 2026-09-21
- Battery: v1
- Scoring: reading 30 pts · surface 50 pts
- Docs: https://docs.roboflow.com/

Three AI models, GPT 5.6 Sol, Claude Opus 5, and DeepSeek v4 Flash, each read Roboflow’s public documentation independently and attempted five first-hour developer jobs: train the first model, find the exact limits, recover from API errors, verify a workflow webhook, use the Python 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.

## Freshness

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

- Category: [AI models & inference](https://docsforagents.com/grades/?category=ai-models)
- Tested: 2026-08-19
- Quotes verified: 2026-08-19
- Surface rechecked: 2026-09-21

No change since the test.

4 of 5 quoted passages still appear on the live pages. The quoted page changed for Find the exact limits.

## Agent surface checks · 40/50

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

## The Reading Test · 21/30

| Task | GPT 5.6 Sol | Opus 5 | DeepSeek v4F | Consensus |
| --- | --- | --- | --- | --- |
| Train the first model | PASS | PASS | PASS | PASS |
| Find the exact limits | PARTIAL | FAIL | PARTIAL | PARTIAL |
| Recover from API errors | PARTIAL | PASS | PASS | PASS |
| Verify a workflow webhook | PARTIAL | PARTIAL | FAIL | PARTIAL |
| Use the Python SDK | PASS | PASS | PASS | PASS |

Docs platform: GitBook (unscored) · verified 2026-08-19

## What to fix first

These 3 fixes could add up to 18 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 · MCP server · check failed**

   **Found:** Initialize returned an unauthorized response, not a valid MCP initialize result.

   **Fix:** Run a public MCP server for the docs that answers a JSON-RPC initialize call and offers a docs search tool.

   **Evidence:** [mcp.roboflow.com/mcp](https://mcp.roboflow.com/mcp)

2. **+4 points · Find the exact limits · PARTIAL**

   **Found:** The docs send readers to roboflow.com/pricing for plan limits and never state the monthly Hosted API inference quota.

   **Fix:** Add a docs table with each plan's included credits, project and member limits, and Hosted API inference quota.

   **Evidence:** [docs.roboflow.com/platform/billing-and-plans/plans](https://docs.roboflow.com/platform/billing-and-plans/plans)

3. **+4 points · Verify a workflow webhook · PARTIAL**

   **Found:** The Webhook Sink page covers the JSON payload and delivery outputs, but no page shows a remote endpoint receiving it.

   **Fix:** Add a walkthrough that sends a Workflow payload to a remote endpoint, confirms receipt, and checks the block outputs.

   **Evidence:** [docs.roboflow.com/workflows/blocks/blocks/data-storage/webhook-sink](https://docs.roboflow.com/workflows/blocks/blocks/data-storage/webhook-sink)

## What the docs get right

- **Train the first model: 3 PASS votes.** One guide covers project creation, upload, Auto Label review, versioning, training, deployment, and visual testing.
- **Use the Python SDK: 3 PASS votes.** The SDK quickstart and dataset guide use the same authenticated workspace-project-version-download path for a public Universe dataset.
- **4 of 5 agent surface checks.** Present: llms.txt, llms-full.txt, markdown mirrors, docs AI. Missing: an MCP server.

## Train the first model

**PASS**

PASS consensus from 3 PASS.

One guide covers project creation, upload, Auto Label review, versioning, training, deployment, and visual testing. The Model Training guide runs a single unambiguous seven-step path from account creation through project, upload, Auto Label, review, version, train, and deploy, and each step links to a detailed page that documents the web app, HTTP API, Python SDK, CLI, and MCP Server routes for the same operation. A single page walks through every step from zero (create project, upload video, auto-label, review, generate version, train, deploy) with no gaps.

## Find the exact limits

**PARTIAL**

PARTIAL consensus from 2 PARTIAL, 1 FAIL.

Public has 15 credits, 10 projects, and 2 members; Core starts at 30 credits, 20 projects, and 3 members, while Hosted API capacity depends on shared credit use. No page in the documentation states included credits, project limits, member limits, or Hosted API inference quotas for the Public or Core plans; every such figure is deferred to roboflow.com/pricing, and the only concrete numbers in the corpus belong to the Premium Trial and the Research Plan, neither of which is Public or Core. The docs site links to the marketing pricing page for plan limits but does not host the comparison table itself. Hosted API inference quotas are not stated in simple request counts; they are derived from the credits-per-inference-second formula, which requires calculation.

## Recover from API errors

**PASS**

PASS consensus from 2 PASS, 1 PARTIAL.

The docs map all four responses to actions, but they omit idempotency and duplicate-side-effect risks for retried writes. A dedicated Errors and Status Codes page covers 401, 402, 429, and 5xx with a cross-tool mapping table for REST status, SDK exception, and CLI exit code, and the Production Readiness Checklist adds an explicit retry-or-not column stating that 429 and 5xx are safe to retry with exponential backoff and jitter while 400, 401, 403, and 404 must never be retried automatically. Two dedicated pages (Production Readiness Checklist and Errors and Status Codes) comprehensively cover every status code, explicitly label safe vs unsafe retries, and provide cross-tool mapping.

## Verify a workflow webhook

**PARTIAL**

PARTIAL consensus from 2 PARTIAL, 1 FAIL.

The block and test pages cover synchronous mode, JSON payloads, outputs, and sink enabling, but not one complete remote receipt check. The block reference documents synchronous operation via fire_and_forget=False, JSON body construction via json_payload, and the error_status, throttling_status, and message outputs used to verify delivery, but the page ships an unfilled template placeholder in its Common Use Cases section, its Example JSON definition leaves the two most important fields as "url": "<block_does_not_provide_example>" and "method": "<block_does_not_provide_example>", its example json_payload_operations keys on "predictions" while the json_payload it operates on keys on "field", and no page walks through receiving the payload on a remote endpoint. The Webhook Sink block's dedicated documentation page returns HTTP 404. No other page on docs.roboflow.com documents how to configure a synchronous webhook sink, send Workflow JSON, receive it remotely, and verify delivery from block outputs.

## Use the Python SDK

**PASS**

PASS consensus from 3 PASS.

The SDK quickstart and dataset guide use the same authenticated workspace-project-version-download path for a public Universe dataset. Installation states pip install roboflow and Python >=3.10, authentication documents the environment variable, explicit argument, and interactive login routes, and the first-script example downloads a real public Universe version with rf.workspace("roboflow-100").project("poker-cards-cxcvz"), the identical example that reappears on the Download a Dataset page alongside full Version.download() parameter documentation. Installation (pip install roboflow, Python >=3.10), authentication (env var, explicit key, or interactive login), and dataset download from a public Universe project are documented in a single page with working code examples.

## The receipt

> The roboflow.com/credits page mentions that 1 credit corresponds to 500 seconds of inference time. A more accurate formula is the following:

Public has 15 credits, 10 projects, and 2 members; Core starts at 30 credits, 20 projects, and 3 members, while Hosted API capacity depends on shared credit use.

- [docs.roboflow.com/deployment/roboflow-cloud/serverless-api/pricing](https://docs.roboflow.com/deployment/roboflow-cloud/serverless-api/pricing)

## Agent surface notes

Initialize returned an unauthorized response, not a valid MCP initialize result.

The live docs expose a public Kapa assistant through the Ask or search control.

## 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 2026-08-19.

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

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