dondai44423/donsetch

DonSeTch

The web, for AI agents. V5 on works: major change incoming

Why · Demo · Sponsors · Install · Quickstart · The 4 tools · Fetch · Search · PDF · Stealth · Compare · Limits · CLI

DonSeTch gives any AI agent full web research from a single local process: fetch, search, crawl, screenshot. Four tools, zero API keys, zero accounts, Rust, one binary. The transport is built from scratch (no hyper, no Playwright, no Selenium), so the fetch tier is fast and stealthy, and the tool surface stays small enough to fit an agent's context.

Works with every MCP client (Claude Code, Cursor, OpenCode, Pi, Hermes) and as a standalone CLI.

🤝 Use Bright Data? Support DonSeTch. Their proxy, SERP and Web Unlocker products come through this partner link and part of it comes back to keep DonSeTch free. You pay nothing extra: https://get.brightdata.com/ivqwoicrrlbr Bright Data plugs into the tool itself: the bd SERP provider, the Web Unlocker tier-3 bypass and the unlocker key type. Bright Data setup guide: zones, API token, HTML/JSON and troubleshooting.

(30-second walkthrough: search, bot-wall bypass, crawl)

(Pi agent session: live research with DonSeTch as a native extension)

DonSeTch is free and open source, and stays that way. Sponsorship pays for the time it takes to keep shipping.

If your product is part of this space (proxy platforms, search infrastructure, BYO providers, anything a DonSeTch user would plug in), Gold goes one step further: fit natively inside DonSeTch and you get the placement plus an official integration shipped in the binary itself.

How to sponsor. Email bhandaribishesh879@gmail.com, or use the ♥ Sponsor button on this repo. GitHub Sponsors payouts aren't available in my region yet, so email is the reliable route: write me, I'll share my PayPal, and we go through the steps. One-time sponsorships are welcome at any amount.

Prepay to lock your rate. You can pay for any number of months in one go at today's price, and that rate is fixed for the whole term, so 6 months of Gold at today's $49/mo is $294, and a later increase (even to $150/mo) never touches it. Price changes only ever apply to sponsorships started after them.

npm, any platform (recommended):

npm install -g donsetch

Downloads the prebuilt binary for your platform from GitHub Releases with SHA256 verification. No build tools needed.

Homebrew (macOS/Linux):

brew tap dondai44423/donsetch && brew install donsetch

Pi agent (native extension):

pi install npm:donsetch

Registers the tools as native pi tools, spawns the binary at session start, self-updates with pi update --extensions.

DeepSeek Harness (dsh, first-class plugin):

dsh plugin --profile web add github:dondai44423/donsetch-dsh

One line and every dsh agent gets fetch, search and crawl as native donsetch_* tools, no mcp__ names and no manual MCP config: the plugin downloads the verified binary for your platform, registers the tools in-process, auto-updates with DonSeTch releases, and picks up donsetch keys add changes live. See the donsetch-dsh repo for the config reference.

Homebrew and dsh auto-track published releases.

Verify the install:

donsetch doctor          # grouped health checks, including a real browser launch
donsetch doctor --deep   # adds live egress, provider and captive-portal probes
donsetch doctor --fix    # repairs mechanical problems automatically
donsetch doctor --json   # appends machine-readable check results
donsetch doctor --mcp    # detects clients and prints absolute-path registration blocks

git clone https://github.com/dondai44423/donsetch.git
cd donsetch
cargo build --release --features ocr,rerank,http

sudo apt-get install -y cmake build-essential pkg-config libclang-dev clang lld nasm golang-go
RUSTFLAGS="-C link-arg=-fuse-ld=lld" cargo build --release

DONSETCH_BROWSER_BACKEND=chromium donsetch doctor             # default backend
DONSETCH_BROWSER_BACKEND=headless donsetch doctor --deep      # forced headless
DONSETCH_BROWSER_BACKEND=cloakbrowser \
  CLOAKBROWSER_BINARY_PATH=/path/to/chrome donsetch doctor --deep

MCP server (for any agent). Register and go:

{
  "mcpServers": {
    "donsetch": {
      "command": "donsetch",
      "args": ["mcp", "--supervised"]
    }
  }
}

--supervised is the crash-only daemon: a panic is a blip, the daemon restarts itself, the session survives. Without a global install, use "command": "npx", "args": ["donsetch", "mcp"].

HTTP transport instead of stdio: donsetch mcp --http --port 8765, clients connect to http://localhost:8765/mcp. Sessions, cancellation, /health, token auth via DONSETCH_HTTP_TOKEN and per-request timeouts are documented in donsetch mcp --help. A request that carries a browser Origin is refused unless the page is on loopback or CORS mode (DONSETCH_HTTP_CORS, which requires the token) is on, and a server bound to loopback answers only to loopback Host names on its own port, so a web page cannot drive it and a DNS-rebound name cannot read from it.

CLI (for humans and scripts). Same engine, thin adapter:

donsetch fetch https://example.com --focus "pricing"
donsetch search "rust async patterns" --intent code
donsetch crawl https://docs.python.org --mode map --topic asyncio

const text = typeof result === "string"
  ? result
  : result.content.filter(block => block.type === "text")
      .map(block => block.text).join("\n\n");
const end = text.indexOf("\n");
if (!text.startsWith("[meta] ") || end < 0) throw new Error("Missing result state");
const state = JSON.parse(text.slice(7, end));
const markdown = text.slice(end + 1).trimStart();
// Inspect state.ok, content_ok, next_offset, resume or next_action.

Tool schemas: donsetch tools (same JSON as MCP tools/list). Every failure is structured: a stable code (wall.challenge, guard.ssrf, deadline.hit, network.dns…), an errorKind (permanent, transient, walled), and a next_action line, so agents branch on codes instead of parsing prose. The model surface carries evidence and the state that changes the next action; transport telemetry (tier, quality, escalation trace, timings, engines) stays under _meta, for example _meta["com.donsetch/fetch-debug"].

Plain HTTP first, ~100-300ms. Wall or JS shell detected, auto-escalate to the ghost browser, solve, bounce the cookies back, refetch at full speed.

DonSift extraction: HTML bytes in, agent-native markdown out. Typed blocks (heading, paragraph, list, table, code, quote, media) with heading breadcrumbs.

• focus: BM25-relevant passages with heading context. Scores use each passage's own text; a repeated page title or common subject cannot make every paragraph relevant. Tables keep selected rows with their headers and complete cells; unordered lists keep selected items with their nested conditions. Omitted blocks, rows and items are counted. Lexical hits stay fast; no match or less than 20% saving returns full content with a notice. 12-language BM25: CJK unigrams and bigrams, stopword lists, stemming, accent folding.
• selector: extract only matching HTML subtrees. A miss returns selector.nomatch with the inspected element count; invalid syntax returns selector.invalid. Non-HTML responses cannot satisfy a CSS scope.
• toc + section: see the outline first, then target one section. Two cheap calls instead of one expensive one.
• Token policy: links stripped by default (~30% off), link farms and wiki junk dropped, duplicates suppressed.
• Classification: Article / Listing / Forum / Docs / Table / Page, a 0-1 quality score, and inline trust signals (focus-miss, JS-shell warning, empty content).
• Page memory: every fetch is fingerprinted, so a re-fetch reports changed with section-level diffs, and since_last=true collapses a re-check to one line (~30 tokens).
• must_contain: verifies a claim against the full page but returns MATCH/NO-MATCH plus up to 3 excerpts (~60 tokens instead of 4k).
• stitch=true: collects up to six same-host parts within 1 MiB. Output obeys max_chars; next_offset resumes collected text and next_part identifies an unfinished part.
• Read budgets: mode=scan / read / deep set 800 / 4000 / 16000 extracted bytes; explicit max_chars wins. The source title/URL header and structured metadata sit outside this body budget. read_status and content_complete distinguish partial output from a complete read. Reddit subsets expose partial and known item counts.
• deadline_ms everywhere: real MCP cancellation, progress notifications, ms cost footer. Nothing can silently hang.
• API/JSON endpoints: .json URLs and JSON responses are terminal at HTTP at any tier; for other API paths pin tier:1 to keep the read on plain HTTP.
• Domain adapters: Reddit, npm/PyPI/crates.io/Go/RubyGems, GitHub, Stack Overflow, Wikipedia and docs sites get restructured from each site's own keyless surfaces. Labeled via=adapter:…, kill-switchable.
• Anti-cloak check: on decoy-prone domains, tier-1 responses are equivalence-checked against a headless render, so decoy suspected is stamped instead of silently passing as content.

read_status describes the returned projection: content has no remaining rendered bytes, partial has withheld bytes or a source subset, probe is a MATCH/NO-MATCH answer, unchanged is a history comparison, and thin is a suspected shell. Failures use walled, notfound, or error. Always pair it with content_complete: a projection can be exhausted while the source is incomplete. next_offset is a byte offset into collected markdown, not a page number. links and media share the rendered markdown budget, including their markup; budget_scope makes that accounting explicit.

For parallel reads, pass an array to web_fetch: successful siblings remain available beside individual error rows. Classified failures return normally with ok:false, code, and next_action; protocol errors remain JSON-RPC errors. Inspect each result before using its evidence.

shot always returns a receipt (requested, saved_to, reason), including when no captcha capture was needed. web_screenshot returns the inline PNG and a private local path containing the same bytes; remote clients should use the inline image because the path belongs to the server host.

Wall time and freshness (observed). Plain HTTP reads 0.3-2s; search 0.7-2.9s; guards are instant (an SSRF block ~13ms, a solve-cooldown refusal ~33ms); auth-wall refusals 11-14s; a browser escalation 17-31s; an interactive-captcha give-up 41-42s (the suite worst case). Dead URLs return the live failure directly; automatic archive recovery has been removed. A readable URL path may supply a suggested_query for locating a moved copy; query parameters and fragments are excluded. Repeat calls are cache-backed: a repeated search returned in ~10ms against ~5.0s cold, a re-fetched page 1.5s against ~11.9s cold. changed and since_last verdicts compare stored history, not a live revalidation: fast and fresh are different axes.

A worked reading ladder (approximate costs): search, then toc (~170 tokens for the outline), then section on the right part (~840 tokens, a complete read), or must_contain for a yes/no (~70 tokens). focus is a convenience reducer; toc -> section is the guaranteed one. Truncated output resumes with next_offset (same URL, offset verbatim); a stopped crawl resumes with resume alone.

Stack Overflow question reads use the public question API with answers and comments. Empty, failed or incompatible API responses retain the website fallback; service-declared backoff is honored. Deleted/private questions and login-only dashboards require their own access. Fetch keeps its direct-IP default even with a configured proxy pool; search and crawl retain their pool routing.

Crawl caps apply to each page's rendered markdown and the aggregate page content. Headers and inventory metadata are separate. A capped page carries next_offset and content_complete: false; it never becomes a full-content history baseline. Full crawl output summarizes the unfetched inventory; structured map retains the URLs. Map mode reports mode: "map", discovered, and InventoryComplete when finished, without an empty content-page array. Map is sitemap-backed when the site has one; without a sitemap it falls back to the seed page's links. Continue a stopped crawl using resume alone; a new crawl requires url.

Search reports requested_results, returned_results, and underfilled when engines cannot supply the requested count. weak considers lexical query coverage as well as engine agreement; it is a retrieval warning, not a fact-check. source_type identifies video, community, paper, reference, publisher, official, institutional, and general web sources from known hosts. It does not certify a page's claims or assign trust to lookalike domains.

The diagnostic domain-profile route cold means ordinary HTTP without verified bot-wall clearance cookies. Repeated successful reads can stay on that route while reusing connections; it does not mean transport caching or learning failed.

Tier 3 bypass (opt-in). When the ghost itself hits a hard wall, fetch falls back to Bright Data Web Unlocker if a key is configured (donsetch keys add unlocker "<token>::<Web Unlocker zone>"). It returns HTML into the normal extraction pipeline; a challenge or empty shell remains a failed fetch. Failures carry recovery guidance on the escalation trace. donsetch doctor --deep checks token access, zone name and product type for free; tokens without account-zone read permission report an unverified check. Successful responses are cached locally (URL and render-mode keyed, sliding 6h TTL, 200 entries); concurrent calls within one daemon share the cache. The daily cap counts every API attempt, including retries, and the whole operation shares one timeout. A timed-out paid request is not replayed automatically. Settings live in [bypass]: donsetch config show. See the setup guide. DonSeTch works without it.

Same-domain, best-first. Two phases: sitemap discovery (cheap URL inventory), then a Governor-paced frontier walk with extraction per page.

• Modes: full (map + content), map (URL inventory only), content (BFS from the seed, no sitemap).
• Focus-ranked frontier: focus="query" ranks pages by BM25 relevance over link text and URL path, and crawls only matches. No semantic matching before fetch: a link sharing no token with the query is never enqueued.
• Adaptive pacing: the Governor paces per (host, lane). 429/503 back off on host signals, host-declared waits (Retry-After, robots Crawl-delay) are honored in full, and everything self-inferred caps at ~7s. Zero artificial dwell on the fetch path: stealth through truth, never time.
• Crawl-shape: frontier pops get a seeded reader-like jitter so repeated crawls never replay one identical, score-eager order into server logs. Ordering only, payloads untouched. Kill switch: DONSETCH_NO_CRAWL_SHAPE=1.
• Resume tokens: a stopped crawl resumes in one call, valid 30 minutes, survives restarts.
• Near-duplicate detection: title plus first 200 chars, hashed.
• Budgets: per_page_max and max_total_chars honor positive character limits down to 1, including concurrent workers. deadline_s accepts 1-600 seconds and bounds discovery, page reads, browser handoffs and extraction. A sliced page carries next_offset; use web_fetch with that offset to continue it.
• Inventory: map_limit controls returned URLs (default 120, up to 50,000). inventory.complete=false discloses output caps, unavailable sitemaps and discovery safety bounds (100,000 URLs, 32 sitemap reads). A map derived from seed-page links cannot establish site-wide completeness. Narrow focus/include_paths for large sites.
• Change tracking: since_last=true rechecks sources, then suppresses unchanged pages using per-URL fingerprints from earlier fetches or crawls. Pagination does not prevent recording a full-source fingerprint; partial output never becomes full baseline text.
• Focus receipts: page state records whether passages were omitted and the block counts. Dataset JSON Lines use dataset_version:2, adding the per-row focus receipt.
• Honest stop reasons: FrontierEmpty, MaxPages, CharBudget, DepthLimit, Deadline, ThrottledOut, Cancelled, InventoryLimit. A drained frontier is complete only when the requested evidence was obtained without failures or excluded pages.
• Cross-process politeness: one host-pace store shared by every crawler on the box, so two runs against one site do not double the rate.

No API key, no account. Six keyless engines across four independent index families, plus eight official verticals, run in parallel on your machine, then merge, dedupe and rank.

• Backends: Bing family (Bing, DuckDuckGo, Yahoo), Brave, Mojeek, Google, plus keyless verticals (GitHub, Wikipedia, HN, Semantic Scholar, arXiv, StackExchange, MDN, Google News).
• Native Google, browser-free: HTTP through the legacy mobile endpoint on the existing Rust transport, with seven experimentally verified Nokia profiles (6230-03.15 by default). A profile that succeeds stays preferred per egress, CAPTCHA advances the cursor circularly, rate limits do not. Retry, pacing and quarantine use the same policy as every other engine. Availability depends on Google and your network; see configuration and limits.
• Ghost SERP cascade: if the fan-out and its retry wave leave the merge thin (under 3 lanes or under 15 hits) and native Google failed, one headless render can recover Google's desktop SERP. It costs nothing when healthy, only fires under underdelivery, and reports itself honestly as google_ghost.
• Semantic reranking: a local ONNX cross-encoder (ms-marco-MiniLM-L-6-v2, 23MB) reads query + title + snippet through full attention, blended 60/40 with RRF + BM25 + consensus. A post-enrichment top-up re-scores close calls using the destination page's real title and description instead of SERP fragments.
• Consensus: a URL several independent indexes return gets a boost. Every result carries score, consensus (independent index families) and engines, and the compact surface shows · N sources.
• Learned engine health: per-engine trust EWMAs and chronic-failure quarantine survive restarts, so dead engines get benched instead of burning fan-out slots on every query.
• Entity coverage: anchor entities (B-tree, version numbers, years) are checked against results. Wrong entity, 0.3x.
• Honest reporting: weak=true means low consensus, per-engine status is always visible, and there is no fake "no results".
• Warm handoff: search pre-fetches the top results, so the next fetch S1 serves from cache in ~3ms. It also pre-solves: if the top result's domain is a known wall, a bounded background solve starts while the agent reads.
• Query compiler: site:, filetype: and intitle: are sent only to engines that honor them, and SERP instant answers are extracted with their source URLs.

Keyless quality, 110 questions across 11 niches with no keys: 95.5% answer-in-snippet against Tavily's published 93.3% (LLM-graded). Reproduce with python3 bench/search_quality.py --verbose; methodology and per-niche numbers live in the script. Search-quality caching for the bench lives in ~/.cache/donsetch/bench-search/, delete it when comparing binaries.

Paid providers add rate limits and premium sources on top of the keyless chain.

donsetch keys add tavily tvly-...       # Tavily
donsetch keys add exa sk-exa-...        # Exa (stackable)
donsetch keys add serper ...            # Serper.dev
donsetch keys add serpapi ...           # SerpApi
donsetch keys add serpbase sb-...       # SerpBase Google SERP (100 free searches)
donsetch keys add bravesearch ...       # Brave Search API
donsetch keys add tinyfish sk-...       # TinyFish (free tier)
donsetch keys add parallel nKil3...     # Parallel AI (fast mode)
donsetch keys add bd "<token>::<SERP zone>"           # Bright Data SERP
donsetch keys add unlocker "<token>::<Unlocker zone>" # Bright Data Web Unlocker
donsetch keys default local             # dispatch order: keyless first

• Stack keys per provider: two Exa keys are one 3,000-credit pool. donsetch keys export/import moves the store.
• Automatic fallback to keyless when a provider errors or runs dry, per-key rate-limit cooldown and depletion tracking.
• The native adapter list is deliberately short, only the biggest services get one. Everything else belongs in the plugin system below, wired by you, running in the same chain.
• Bright Data SERP, Web Unlocker and the proxy/data products are available at get.brightdata.com (affiliate link).

Not natively supported is not the same as unsupported. Register any executable that answers a small stdin/stdout JSON contract and DonSeTch treats it like any other provider: default chain, fallback, attribution. Any language works.

donsetch keys add plugin searxng --cmd 'python3 ~/searxng-adapter.py' --test

Request on stdin, response on stdout:

{"format":1,"query":"rust async","max_results":8,"intent":"web","deadline_ms":30000}
{"format":1,"results":[{"title":"...","url":"https://...","snippet":"...","score":0.9}]}

Errors: exit non-zero with a message on stderr, or answer {"format":1,"error":"...","retryable":true,"error_kind":"rate_limited"}. error_kind is optional and says what actually went wrong: invalid_key, credit_depleted, rate_limited, server_error, network_error. Report one of the first three and the plugin is parked the way a native key in that state is parked, rate-limited for a 60s cooldown then retried, invalid or depleted until you re-register it with keys add plugin, and keys list shows which. Leave the field out and nothing changes: retryable alone still only picks between retrying this search elsewhere and reporting the failure. Constraints that keep it reliable: hard timeout (default 30s, --timeout to change), 8 MiB stdout cap, direct exec with no shell, killed on cancellation, and a malformed response degrades gracefully down the fallback chain. Keys belong in the adapter's own environment, never in DonSeTch config.

PDFs are detected by Content-Type or %PDF magic and parsed through a custom PDFium FFI. No external PDF library, no Python subprocess.

Glyphs and rendered pixels come from the same content stream, so they are already aligned. Pixels tell the truth about structure, glyphs tell the truth about text, and the fusion is deterministic: no hallucination.

Tier B, lazy: OCR through the PP-OCR cascade (En → Zh → Deva) for scans and broken ToUnicode pages.

web_fetch takes an actions array, executed in the real browser before extraction:

{
  "url": "https://duckduckgo.com",
  "actions": [
    { "do": "type", "selector": "input[name=q]", "text": "rust async tokio" },
    { "do": "press", "key": "Enter" },
    { "do": "wait_text", "text": "tokio" }
  ],
  "focus": "tokio"
}

Steps: wait, wait_selector, wait_text, click, hover, type (human cadence), press, scroll. Up to 16 steps, deterministic waits, per-step results in structuredContent.actions. Search flows, form submits, load-more buttons: one call, no separate browser tool.

Everyone else patches a foreign TLS stack to resemble Chrome and ships hardcoded fingerprint tables that rot. DonSeTch drives Chrome's own BoringSSL with Chrome's native behaviors on: GREASE, extension permutation, ECH-GREASE, ALPS, SCT, OCSP, cert compression. The ClientHello comes out of the same machinery Chrome uses.

Own HTTP/2 stack. Off-the-shelf h2 does not expose pseudo-header order, the exact SETTINGS set, WINDOW_UPDATE values or HPACK indexing strategy, all of which are fingerprintable. So DonSeTch has its own: full HPACK (257 Huffman symbols, 61 static entries), frame engine (SETTINGS, DATA, WINDOW_UPDATE, PING, GOAWAY, RST_STREAM, CONTINUATION, PRIORITY), request HEADERS carrying Chromium's priority flag and block, flow control with replenishment, TLS 1.3 session resumption, connection pool. The h2 preface is asserted byte-identical to Chromium in CI, so a detectability regression is a build failure.

Temporal stealth. The part that is not in the bytes:

One identity per fetch. Per-domain personas drive Accept-Language, the ghost viewport and navigator.languages, so tier 1 and the browser claim the same person. The profile is derived from the browser actually installed, and donsetch --version tells you which one it detected.

The browser almost never fetches content. It exists for the two things HTTP cannot do: pass a JS challenge and execute a JS-rendered page. Its output is cookies (handed to tier 1, which then fetches at full speed) or rendered HTML (handed to the extraction engine).

Raw CDP launch without automation flags, so navigator.webdriver is natively false. Nothing is injected before page scripts: a real window, real GPU, real locale, a real PluginArray, and the window.chrome object Chrome actually ships. Where a stealth plugin patches a property so it looks less automated, this leaves the real one alone, because a patched property is a fingerprint of its own.

Every fetch is an action and an observation. Pure deterministic state, no ML, and the loop converges: the more you use it, the less it escalates.

Cookie lifetime converges: observed_lifetime = min(previous, now - last_solved). Only clearance cookies persist (cf_clearance, datadome, _abck); tracking cookies are filtered out.

A wall that survives two real-browser solves is remembered, so future fetches answer honestly in milliseconds until the backoff lapses. A single failure never gates a domain, and any real solve or tier-1 cold success clears the memory. The daemon learns what its own environment cannot do.

Per-intent engine trust, egress-class domain profiles (cookies never cross exits), domain quality priors and crawl governor ladders all feed the same loop, all local, with receipts in donsetch status and donsetch doctor --improve. Disable disk state with DONSEEK_NO_DISK_STATE=1.

Every layer in Rust, no dependency on existing OSS web tooling.

1,300+ tests. Zero clippy warnings: cargo clippy --all-targets --features ocr,rerank -- -Dwarnings is the law, and CI runs the full matrix on Linux, macOS and Windows.

Every runtime knob lives in one typed config (src/config.rs), layered, later wins:

1. Compiled defaults. A bare donsetch mcp stays the law: zero config needed.
2. Legacy env vars (the pre-v4 names, DONSETCH_NO_CRAWL_SHAPE say). Honored exactly as before, reported as deprecated by doctor.
3. donsetch.toml at <config-dir>/donsetch/donsetch.toml, or anywhere via DONSETCH_CONFIG=/path/file.toml. Unknown keys and bad values are hard errors naming the file. DONSETCH_NO_CONFIG_FILE=1 skips the file layer (setting both is an error).
4. New env names: DONSETCH_<SECTION>__<KEY>, so DONSETCH_FETCH__PDF_MAX_MB=25.

Sections: transport, mcp, paths, state, proxy, tls, persona, cli, fetch, bypass, search, browser, debug, rules.

[fetch]
h3 = true               # opt into the h3 lane
shadow_fetch = "never"  # or "auto" / "always"
pdf_max_mb = 100
dns_cache_ttl_secs = 30 # seconds a resolved name is reused in-process, 0 disables

[bypass]
max_daily = 100

Booleans take true/false, enums take their listed spellings.

• donsetch config show prints every knob with its value and its origin (default, legacy, file, env).
• donsetch config show --markdown prints the full reference table, one row per knob.
• donsetch config show --legacy maps every old env name to its config key.
• donsetch doctor warns about the legacy vars active in your shell, mapped to their keys.

Some sites treat any automated traffic as hostile, however well it is disguised. The block lands on your IP, sometimes on your whole network, outlasts the session that caused it, and also locks you out in your own browser. A note in the agent's instructions file ("never fetch site X") helps only if the model reads it and obeys it every time, and one forgotten turn is enough to get banned. Other sites need a particular tier every time, and the agent has no way to know that in advance. Pick the right tool covers which task shapes draw these blocks. Local rules enforce your answer inside DonSeTch, whatever the model decides.

Per-host rules in donsetch.toml can refuse a URL before any request, handing the agent your own instructions in place of the page, or pin the fetch tier for a host. DonSeTch ships none; the table is empty until you write one.

[rules.url."example.org"]        # the host and every subdomain, http and https
action  = "deny"
reason  = "ip_ban"               # error code policy.denied.ip_ban
message = "do not fetch this site with curl, scripts or another HTTP client: it bans our IP for automated access. Find file elsewhere, or ask the human operator to download it"

web_fetch, web_crawl and web_screenshot then refuse that site, redirects and cached copies included, and the agent reads message as its Next action:, so lead with what to do instead, and say why when the reason rules out workarounds. The most specific matching rule wins in full: a narrower rule that only sets tier also cancels a broader deny. donsetch rules test <url> shows which rule a URL gets and why, offline.

Rules are a guardrail, not a security boundary: no MCP argument can switch them off, but an agent with a shell can run donsetch fetch --ignore-rules, set DONSETCH_RULES__MODE=off, or edit donsetch.toml. Pattern grammar, load errors, precedence, tier pins, what the agent sees and the known limits: local rules.

Thin adapter over the same engine the MCP server uses.

Every management command validates its flags: an unknown one is refused with the nearest known match and exits 1. A usage error is never a silent run. The proxy help and its empty-pool state carry one labelled sponsor line (Proxy-Seller); DONSETCH_NO_SPONSOR=1 removes it, and it never appears in --json, proxy export, or any MCP result or tool definition.

Adding a proxy is one command. Nothing else is required.

donsetch proxy add socks5://user:pass@1.2.3.4:1080   # validated, probed, then saved
donsetch proxy list                                  # what is configured (credentials masked)
donsetch proxy check                                 # live probe: latency + real exit IP

• Formats: http://host:port (HTTP CONNECT, credentials optional as user:pass@), socks5://host:port and its socks5h:// alias (SOCKS5 sends the target as a name, so the proxy resolves DNS and your resolver never sees the host). A bare host:port is HTTP CONNECT. Anything else is refused with the reason.
• Many at once: donsetch proxy add url1 url2 url3, or a whole file with donsetch proxy import proxies.txt (one URL per line, # comments) — donsetch proxy export writes that file back out. --no-check skips the probe when you are adding hundreds.
• Try before you add: donsetch proxy test <url> probes one proxy without saving it.
• Where they live: donsetch proxy add appends <cache-dir>/proxies.txt (mode 0600, credentials included). DONSEEK_PROXIES="url1,url2" in the environment and [proxy] pool = ["url1", "url2"] in donsetch.toml are the other two doors; all three merge, and an environment/config entry replaces a file entry with the same host:port. The pool is read at startup, so restart the MCP server to pick up a change.
• Removing: donsetch proxy remove 1 (the index proxy list prints, or host:port) and donsetch proxy clear.
• Checking is a tunnel, not a TCP handshake: the probe connects through the proxy to api.ipify.org and prints the exit IP, so a proxy that accepts the connection and then cannot carry traffic is reported dead. donsetch status shows the count, donsetch doctor shows every lane and its health.

A proxy you configure is a proxy that sees your traffic, so nothing uses one by accident.

donsetch proxy fetch on makes web_fetch (and its browser escalation) sticky to a pool lane per host, rotating on 429/407/dead/timeout. A one-shot read does not need a shared exit, which is why fetch is opt-in while search and crawl are not.

Crawl is where the first request decides things: a host that bans on contact has already seen your own IP before a lane can take over. When a site must never see your address at all, keep it off entirely with a local rule rather than leaning on the pool.

Set the environment and every fetch follows the curl convention:

• HTTPS_PROXY / HTTP_PROXY / ALL_PROXY (any case) route fetches through that proxy, and NO_PROXY exempts hosts (suffix match, *, host:port, CIDR). DONSETCH_NO_ENV_PROXY=1 disables the convention entirely. [proxy] http / https / all in donsetch.toml do the same thing without environment variables.
• A configured proxy is a promise: if it cannot be dialled the fetch fails and says so, instead of quietly leaving on your own IP. Same for a pool lane: a dead lane is benched, not replaced by a direct read of a host it was carrying.
• In an intercepting network (a TLS-terminating middlebox), HTTP CONNECT gets an interception-safe handshake (no GREASE/ALPS/ECH/compress-cert, which some middleboxes reset on) and SSL_CERT_FILE / SSL_CERT_DIR load into the trust store, exactly like curl. SOCKS5 keeps the Chrome-true handshake with TLS end to end. donsetch doctor reports the whole posture and names the interception fix when a live check fails.
• Authenticated lanes work in the browser tier too: Chrome cannot carry proxy credentials, so it is pointed at a local credential-free relay that authenticates upstream, instead of dying as ERR_SOCKS_CONNECTION_FAILED.

Pages behind a login (x.com, gated docs, internal tools) need a real session. donsetch login gives you one without ever seeing your credentials:

donsetch login x.com          # opens YOUR browser, you sign in, press Enter
donsetch login --list         # stored sessions (names and counts only)
donsetch login --status x.com # one domain, in detail
donsetch login --logout x.com # forget a domain
donsetch login --import cookies.txt x.com   # servers and CI, Netscape format

• Credentials never enter DonSeTch. It opens a real Chromium on your display in a dedicated profile, never the automation one. You type into the browser. No keystroke capture, no screenshots, no CDP attach until you press Enter.
• Afterwards, cookies selected by your domain or explicit import are stored regardless of whether their names look like authentication tokens. Known analytics and clearance cookies, expired records and invalid values are excluded. Names and paths retain their identity, including opaque persistent names such as c_user, xs and li_at. Cookie values stay in the private vault; Saved/Imported lists names only. Tier-1 fetches and warm tier-2 browsers reload the session without a daemon restart. Login changes invalidate cached account documents; captures from older browsers cannot overwrite the replacement session.
• A named-domain post-login probe checks reachability and reports login/access redirects as unverified. A public home-page HTTP 200 cannot prove authentication, so it never earns a verified checkmark. Multi-site saves avoid one probe per domain. Verify the actual protected page you need with web_fetch.
• Failed writes and captures return an error instead of reporting success. EOF and Ctrl+C save nothing to the vault; --fresh waits for the exclusive profile lock before wiping it. The dedicated browser profile may retain its own cookies until cleared with --fresh.
• The registry (auth-state.json) stores metadata only: names, counts, expiries, probe verdicts. Never values.
• Multi-site: run bare donsetch login, sign into as many tabs as you like, press Enter once. New cookies and rotated values are captured, including opaque persistent cookies.

docker build -t donsetch-mcp .
docker compose up -d                 # loopback-only by default

Multi-stage build, non-root user, optional Chrome, resource limits in the compose file. An opt-in http profile serves the HTTP transport with a healthcheck, and docker compose stop gives in-flight tier-2 fetches a 45s grace period.

DonSeTch is a rapid-fire research tool: search, read, verify. A search, a fetch or two, a docs page, one PDF. Its speed is the point, and that speed is its stealth for one-shot reads.

It is NOT built for tasks where an agent "works through" a defended site the way a person would:

• Bulk document harvesting: discovering and downloading many PDFs from one repository in a single run.
• Long sessions against one site: page after page, at machine speed, same IP, no human pauses.
• Mass extraction: mirroring a file library, collecting a dataset, systematic downloads.

DonSeTch will probably work on those, and nothing stops it. But every request fires hundreds of times faster than a human, and a defended site reads that pattern itself as a bot, not just the fingerprint. The realistic risk is an IP-level block, sometimes on your whole network mid-run. When that happens it is the task shape, not a fetch-layer failure: the same page fetched once, as a research read, is fine.

For that other shape of work, use Bladebro: a real browser doing what a human does, page by page, download by download, at a human's pace. Slower than DonSeTch by design, because over a long session against a defended site, looking human beats being fast.

Rule of thumb: one-shot research = DonSeTch. Working a defended site like a person to collect things = Bladebro.

The honest summary: nothing else in this space combines a real Chrome TLS stack, browser-free keyless search, PDF pixel-fusion and local-only operation in one binary.

Against Firecrawl, head to head, same tasks, live. Firecrawl is a paid cloud API, DonSeTch is free, local and keyless:

• Fetch: Wikipedia comes back 16x smaller (16KB against 267KB). An arXiv PDF is 22x faster and 4.4x smaller (1.4s/16KB against 32.6s/71KB). Reddit: Firecrawl refuses the site, DonSeTch returns the real feed.
• Crawl: 2.5x faster on the same docs target (18.7s against 47.6s), with --topic ranking the frontier instead of firehosing, and an honest small failure when the topic has no match instead of verbose unrelated content.
• Search: Firecrawl is genuinely faster (1-2s against 5-7s) and leans mainstream authority. DonSeTch leans technical specificity, matched or beat it on exact GitHub issues, and costs far fewer tokens per query.

PRs welcome, see CONTRIBUTING.md. Before submitting: just check, just t <scope> for the area you touched, and just lint (clippy with -Dwarnings). CI runs the full matrix on three platforms. AGPL v3: all contributions land under the same license.

Copyright (c) 2026 Bishesh Bhandari. AGPL-3.0, see LICENSE.