# OpenCode docs: AI Agent Readiness Score 42.5% (F)

**42.5% · 34/80 · AI Agent Readiness Score · 24/30 reading points · 10/50 agent surface points**

OpenCode received 9 PASS votes and passed 1 of five agent surface checks. The clearest finding came from the start the first coding session task.

- Tested: 2026-08-20
- Published: 2026-09-21
- Battery: v1
- Scoring: reading 30 pts · surface 50 pts
- Docs: https://opencode.ai/docs/

Three AI models, GPT 5.6 Sol, Claude Opus 5, and DeepSeek v4 Flash, each read OpenCode’s public documentation independently and attempted five first-hour developer jobs: start the first coding session, find the exact Zen limits, recover from a provider failure, authenticate the OpenCode server, use the JavaScript 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: Provider and session recovery replace HTTP 429. T4: HTTP basic 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.

5 of 5 quoted passages still appear on the live pages.

## Agent surface checks · 10/50

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

## The Reading Test · 24/30

| Task | GPT 5.6 Sol | Opus 5 | DeepSeek v4F | Consensus |
| --- | --- | --- | --- | --- |
| Start the first coding session | PARTIAL | PARTIAL | PASS | PARTIAL |
| Find the exact Zen limits | PASS | PASS | PARTIAL | PASS |
| Recover from a provider failure | PARTIAL | PASS | PASS | PASS |
| Authenticate the OpenCode server | PASS | PASS | PARTIAL | PASS |
| Use the JavaScript SDK | PASS | PARTIAL | PASS | PASS |

Docs platform: Astro Starlight (unscored) · verified 2026-08-20

## What to fix first

These 5 fixes could add up to 42 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.txt · check failed**

   **Found:** The site and docs-root candidates returned HTTP 404.

   **Fix:** Publish llms.txt at the docs root as plain text, with a title, a one-line summary, and a link to every page.

   **Evidence:** [opencode.ai/llms.txt](https://opencode.ai/llms.txt)

2. **+10 points · llms-full.txt · check failed**

   **Found:** The site and docs-root candidates returned HTTP 404.

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

   **Evidence:** [opencode.ai/llms-full.txt](https://opencode.ai/llms-full.txt)

3. **+10 points · MCP server · check failed**

   **Found:** Candidate endpoints returned errors and no valid MCP response.

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

   **Evidence:** [opencode.ai/mcp](https://opencode.ai/mcp)

4. **+10 points · Docs AI · check failed**

   **Found:** The live docs expose standard search but no public assistant control.

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

5. **+2 points · Start the first coding session · PARTIAL**

   **Found:** The intro tells readers to run /connect before it gives the opencode launch command. Opus 5 also found the intro and Providers pages name different /connect entries.

   **Fix:** Give the opencode launch command before the /connect step, and use one /connect entry name on every page.

   **Evidence:** [opencode.ai/docs/index](https://opencode.ai/docs/index)

## What the docs get right

- **Find the exact Zen limits: 2 PASS votes.** The page defines workspace and member limits and warns that auto-reload can make charges exceed a monthly usage limit.
- **Recover from a provider failure: 2 PASS votes.** Named sections cover authentication issues, model-not-available including ProviderModelNotFoundError with the providerId/modelId format rule, ProviderInitError, and AI_APICallError, each with concrete commands, though the ProviderInitError remedy deletes the directory that also holds all session data.
- **1 of 5 agent surface checks.** Present: markdown mirrors. Missing: llms.txt, llms-full.txt, an MCP server, docs AI.

## Start the first coding session

**PARTIAL**

PARTIAL consensus from 1 PASS, 2 PARTIAL.

The guide gives the opencode launch command only after it tells the reader to run /connect in the TUI. The Configure section instructs you to run the TUI slash command /connect before the following Initialize section tells you to run opencode, and three pages name three different entry points for the same step: the intro says select opencode, the providers page says select OpenCode Zen, and the CLI page documents opencode auth login. The docs provide an unambiguous four-step path: install via curl script (or npm/brew/pacman), configure a provider with the /connect command, initialize with /init, then ask a question. No guessing needed.

## Find the exact Zen limits

**PASS**

PASS consensus from 2 PASS, 1 PARTIAL.

The page defines workspace and member limits and warns that auto-reload can make charges exceed a monthly usage limit. The Zen page states the $5 reload threshold, the $20 reload amount, monthly usage limits for the whole workspace and for each member, and the overrun case where auto-reload can charge past a monthly limit. The reload threshold ($5) and reload amount ($20) are clearly stated. Workspace and per-member monthly limits exist. However, the docs contradict themselves on overrun behavior: they first say a monthly limit is absolute ("will not use more than $20") then immediately say auto-reload bypasses it ("might end up charging you more than $20"). An agent cannot determine the actual enforcement behavior.

## Recover from a provider failure

**PASS**

PASS consensus from 2 PASS, 1 PARTIAL.

The page gives concrete recovery commands, but AI_APICallError gets one possible cause and no next diagnostic when cache reset fails. Named sections cover authentication issues, model-not-available including ProviderModelNotFoundError with the providerId/modelId format rule, ProviderInitError, and AI_APICallError, each with concrete commands, though the ProviderInitError remedy deletes the directory that also holds all session data. All four failure types have dedicated diagnostic and correction paths: authentication issues (re-authenticate, check keys/network), ProviderModelNotFoundError (check model name format, run opencode models), ProviderInitError (verify setup, clear config store, re-authenticate), and AI_APICallError (clear provider cache, restart). Complete.

## Authenticate the OpenCode server

**PASS**

PASS consensus from 2 PASS, 1 PARTIAL.

The CLI page documents opencode attach with password and username flags that use the same environment-variable defaults. The server, CLI, and web pages agree on the env var names and the opencode username default, and the CLI page documents opencode attach with --username and --password for the client side, but the JavaScript SDK client options table lists no credential or header field. Server-side setup is clearly documented on /docs/server with environment variables and an example command. Client-side credential handling is documented on /docs/cli (opencode attach and opencode run have --password/--username flags defaulting to the same env vars), but requires cross-referencing two pages. The SDK client (createOpencodeClient) has no documented way to pass HTTP basic auth credentials.

## Use the JavaScript SDK

**PASS**

PASS consensus from 2 PASS, 1 PARTIAL.

The SDK documents npm installation, client creation, session creation, and a prompt body that matches the server API. Install, createOpencodeClient, session.create, and session.prompt are all shown, but the same page contradicts itself twice: the sessions table names body.outputFormat while the structured-output example sends format, and the session example reads session.id while every other example on the page unwraps the documented fields response style through .data. The SDK page documents the complete workflow: npm install, createOpencode or createOpencodeClient, session creation via client.session.create(), and prompt sending via client.session.prompt(). All examples use correct API shapes that match the server's documented OpenAPI endpoints.

## The receipt

> 1. Run the /connect command in the TUI, select opencode, and head to opencode.ai/auth.

The introduction documents provider connection and project initialization, but prerequisites and setup span multiple sections.

- [opencode.ai/docs/index](https://opencode.ai/docs/index)

## Agent surface notes

Candidate endpoints returned errors and no valid MCP response.

The live docs expose standard search but no public assistant 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 5 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/opencode-docs-ai-agent-readiness/