A simpler way
into the web.
Route names, request fields and the technical reference below are in English, as used by the API.
Navigator's HTTP API and MCP tools use one product domain and a Navigator-only identity. Check /v1/capabilities for the functions enabled in the current deployment. Public-page reads and saved results remain separate from HUNTER private operations.
API routes
GET /health- Service health, without internal host details.
GET /v1/capabilities- Machine-readable feature readiness. Trust this instead of marketing copy.
GET /v1/session- Validate a Navigator-only invite key and return its non-secret identity and allowed domains.
POST /v1/fetch- Retrieve a public HTTPS page as bounded structured data. Requires a product identity, or a separately enabled guest trial.
POST /v1/search- Structured search through an independently configured provider.
POST /v1/research- Collect cited excerpts from up to three public sources. Uses Search, Research and page-reading allowances; returns evidence only, with no AI synthesis or independent truth guarantee.
POST /v1/enrich- Copy selected title, headings, public links and excerpt from one public page, with source, reading time and value digests. Uses one Fetch allowance.
GET /v1/runs- Saved public Fetch, Search, Research and Enrich previews. Reading a saved result never repeats the original page request.
POST /v1/datasets- When configured, a human Owner can start a small, durable public-page batch. The request body contains a unique list of up to 10 URLs.
GET /v1/datasets/{id}- Read this connection's job progress and saved attributed rows. Use
GET /v1/datasetsto list jobs. POST /v1/datasets/{id}/cancel- Body
{}. Stop unfinished rows. Requests already sent may still use an allowance. POST /v1/datasets/{id}/resume- Body
{}. A human Owner explicitly retries unfinished rows. Completed rows are kept. GET /v1/datasets/{id}/export- Saved JSON by default;
?format=csvdownloads spreadsheet-safe CSV. Export never reads source pages. GET/POST /v1/agent/connection- When hosted setup is available, a Google Owner can configure a dedicated AI key and daily limit. The key is never returned. Initial limit $0; paid tasks disabled.
POST /v1/agent/runs- Owner-granted public reading goal and starting URL. Dedicated connection and remaining budget required; no private profiles, forms or arbitrary clicks.
GET /v1/agent/runs/{id}- Saved progress, AI interpretation and observed public sources. List with
GET /v1/agent/runs. Saved reads make no model calls. POST /v1/agent/runs/{id}/cancel|resume- Body
{}. Stop future steps or explicitly resume within remaining limits. Already sent requests can cost money. POST /mcp- Navigator-only MCP tools with a separate user identity and scope, never HUNTER Owner operations.
Fetch request
POST https://navigator.hunterall.com/v1/fetch
Authorization: Bearer <your-navigator-access>
Content-Type: application/json
{
"url": "https://example.com/"
}Search request
POST https://navigator.hunterall.com/v1/search
Authorization: Bearer <your-navigator-access>
Content-Type: application/json
{
"query": "nail trends Berlin",
"max_results": 5
}Search appears only after the operator connects a valid server-side SearXNG endpoint and durable Navigator metering. The endpoint cannot be supplied by a caller. Provider egress is bounded by the active deployment contract; meter failure denies before egress. Guest Search remains disabled.
Public dataset request
POST https://navigator.hunterall.com/v1/datasets
Authorization: Bearer <your-navigator-owner-access>
Content-Type: application/json
{"urls":["https://example.com/","https://example.org/"],
"fields":["title","headings","links","excerpt"]}Dataset jobs require dataset: configured, read access and a human Owner's expiring job grant. OAuth AI clients cannot grant themselves background authority. URLs must be public HTTPS without query parameters, fragments or login profiles. Each attempted row uses a normal Fetch allowance; new-job caps start at one per connection, three per tenant and fifteen service-wide per UTC day. Storage caps apply. A crashed row requires explicit Resume; successful rows never replay. The source fields are copied evidence, with empty values for missing data, and do not include AI inference. Jobs and results are retained for 30 days after their last update.
The MCP tool navigator_datasets uses actions create, list, get, cancel, resume and export against the same service, identity and quotas. Job actions use dataset_id; export accepts format: json or csv. Google browser sessions require a same-origin request for state changes; never copy a session cookie into an agent.
Free guest trial
The website workspace can offer a small free Fetch trial. It only works from the Navigator website, never cross-site:
GET /v1/guest-sessionREADYif your browser already has a session, otherwiseCHALLENGE_REQUIREDwith a Cloudflare Turnstile site key.POST /v1/guest-session- Body
{"token":"<turnstile token>"}. On success sets a signed, HttpOnly session cookie valid for 24 hours. 403 CHALLENGE_FAILED- Human check not accepted. Retry the check.
429 SESSION_DAILY_LIMIT / GLOBAL_DAILY_LIMIT- Free quota used up for today (UTC). Use an API identity for more.
503- Verification or metering temporarily unavailable. Navigator fails closed and never runs an unmetered request.
Identity and safety
Tokens are intended to be scoped to Navigator only, with per-tenant usage and origin restrictions. The public launch will require independently verified authentication, durable quotas, private-data handling and external tests. The system never shares its issuer, keys or session state with HUNTER OS.
Feature maturity
Availability is determined by /v1/capabilities and your connection's scopes. A configured function does not establish full-product acceptance or public beta readiness. Dataset execution, Browser, AI Agent, profiles and other optional functions are independently gated; unavailable functions return an explicit error.
Impressum · Datenschutz · Terms · Privacy