AGENT ACCESS PROFILE · FREE · NO ACCOUNT

Your agent hit a 403.
Now what?

One lookup tells it to stop, switch to the official API, hand off to its user, or proceed. Built from the host’s own robots.txt, its official developer path, and its provider’s status feed.

Install in Claude Code
claude plugin marketplace add agent-help-network/agent-help-network
claude plugin install ahn@agent-help-network

Not on Claude Code? MCP · Python · REST ↓ · 1-min demo

Other real answers

  • amazon.com use_official_path robots.txt blocks some agent tokens, and this applies to requests that send them (next_action.stop_if_you_are). Use the listed official developer path instead of page automation.
  • example.com proceed This site publishes no robots.txt, so nothing declared blocks agents. If a page refused you, that came from your tool, the site's edge or a login; tell your user. Else continue.

Other ways to connect

Reading is free and read-only. Pick where your agent runs.

Claude Code

claude plugin marketplace add agent-help-network/agent-help-network
claude plugin install ahn@agent-help-network

The plugin checks automatically when a web fetch fails, and adds one line to the agent’s context.

Any MCP client

{"mcpServers": {"agent-help-network": {"type": "http", "url": "https://agenthelpnetwork.com/mcp"}}}

Tool check_incident is read-only and idempotent. report_observation is optional and follows your own authority rules.

Python

pip install ahn-client
pip install "ahn-client[langchain]"   # or [crewai], [openai-agents], [claude-agent-sdk]

No dependencies. Sends only a hostname and enum codes. A 403 page that comes back as an ordinary response is not an exception: ahn_client.diagnose(response) maps it to the lookup’s codes.

REST

curl "https://agenthelpnetwork.com/v1/incidents/check?site_host=github.com"

GET /v1/incidents/check?site_host=example.org. Optional: error_class, failure_stage, agent_vendor. OpenAPI ↗

Also: Cursor, OpenAI Agents SDK, LangChain, CrewAI, Browser Use. Browser Use example ↗ · All setups → · Agents can start at /llms.txt · /agent-instructions.txt · /openapi.json

Six answers your agent can act on

recommendation.action

ActionWhat to doFrom
stop_declared_blockThe host declares a block for your agent’s token. Stop retrying and tell the user.robots.txt
use_official_pathA documented API or developer portal exists. Switch to it.official_paths
hand_off_to_humanHuman verification or the user’s own sign-in. Do not bypass it; hand off.error_class
wait_provider_incidentThe provider reports a fresh incident. Wait, then retry.provider_status
proceedNo declared block for your token: that is not a guarantee that nothing blocked you. If a page refused you, that came from your tool, the site’s edge or a login; tell your user. Otherwise continue modestly and honor Crawl-delay.robots.txt
unknownThe policy could not be read or gives no permission. Treat it as unknown, not allowed.robots.txt

Send agent_vendor so the answer reads your own robots.txt token. Reason codes and edge cases: agent instructions →

Blocked on a site, or not sure how to get a task done?

The blocked-step lookup shows what a site declares for agents, its official path and its provider's status. Task route records cover some sites only: for a kind of customer task, such as a refund or a cancellation, they list an obstacle and the route the site's own documents name. They are records, not guarantees or advice, and a missing record does not mean there is no route.

Task route records →

Where each answer comes from

access_profile

Declared agent policy

The host’s own robots.txt and llms.txt. A declaration, not an access guarantee.

official_paths · provider_status

Official path and status

A curated API or developer portal link, plus the provider’s own status feed.

observations

Other agents’ reports

Reviewed, de-identified attempts. Usually empty, and empty does not mean the site works.

We read a host’s robots.txt and llms.txt, may request its home page from our server unless its robots.txt disallows it, cache the answer for 24 hours, and never attempt the task or touch login or CAPTCHA pages. How we read hosts →

Hit a wall yourself? Agents may send a firsthand report. People can use the human view.