MCP Server

NSIN hosts a remote Model Context Protocol (MCP) server, so AI assistants can manage your domains, DNS records, cache and edge rules in natural language — “add an A record for staging pointing at 203.0.113.7”, “purge everything under /assets on example.com”, “is the site down right now?”, “how much traffic did my site get this week?”.

Endpoint: https://api.nsin.ir/mcp (Streamable HTTP)
Auth: your API key as Authorization: Bearer nsin_…
Nothing to install — it’s hosted by NSIN.

Connect

Claude Code

claude mcp add --transport http nsin https://api.nsin.ir/mcp \
  --header "Authorization: Bearer nsin_your_key_here"

Claude.ai / Claude Desktop

Add a custom connector in Settings → Connectors with the URL https://api.nsin.ir/mcp, and supply the Authorization: Bearer nsin_… header when prompted.

Cursor

Add to .cursor/mcp.json (or the global ~/.cursor/mcp.json):

{
  "mcpServers": {
    "nsin": {
      "url": "https://api.nsin.ir/mcp",
      "headers": { "Authorization": "Bearer nsin_your_key_here" }
    }
  }
}

Tools

A ✓ in the second column means a read-only API key can use the tool. Read-only keys are served a smaller tool list: the write tools are not just rejected, they are never shown to the assistant.

Domains, DNS and SSL

Tool	Read-only keys	Description
`list_domains`	✓	All domains you can access, with status and your role
`get_domain`	✓	One domain’s full details, nameserver state and settings
`list_dns_records`	✓	DNS records with ids, proxy status, origin scheme/port/host-header
`get_ssl_status`	✓	Certificate validity, expiry, issuer, SANs, coverage gaps
`add_domain`	—	Add a domain to the account (managed or external DNS)
`create_dns_record`	—	Create a record, including origin scheme, port and Host override
`update_dns_record`	—	Change a record’s fields
`delete_dns_record`	—	Remove a record
`check_nameservers`	—	Re-test a pending domain’s delegation and activate it
`issue_ssl_certificate`	—	Order a certificate now instead of waiting for the cycle
`set_developer_mode`	—	Bypass the cache for a few hours while deploying
`update_domain_settings`	—	Cache budget/TTL/body cap, security headers, HSTS, minimum TLS version, Markdown for AI agents

Cache

Tool	Read-only keys	Description
`list_cache_entries`	✓	Browse what is cached: host, path, variant, node, size, expiry
`get_cache_summary`	✓	Total cached objects and bytes, per edge node
`purge_cache`	—	Purge a domain’s entire edge cache
`purge_cache_by_path`	—	Purge (or refresh) everything matching a URI pattern
`purge_cache_entries`	—	Purge (or refresh) exactly the objects you listed

Rules

Tool	Read-only keys	Description
`list_rules`	✓	List one rule kind in evaluation order, with ids and settings
`toggle_rule`	—	Enable or disable a single rule

list_rules and toggle_rule take a kind: cache, waf, drop, rate_limit, captcha, bot_route, fingerprint, redirect, rewrite, header, origin_pool, origin_route, basic_auth, error_page or optimize. Creating and editing rules stays in the panel — the MCP server can read them and park a misbehaving one.

Traffic, logs and health

Tool	Read-only keys	Description
`get_analytics_summary`	✓	Requests, bandwidth, visitors, cache ratio, error rates
`get_top_uris`	✓	Most-requested page URLs over a period
`search_request_logs`	✓	Raw request log, filtered by status, URI, cache, fault, IP, ray id
`list_waf_events`	✓	Firewall events: which rule fired, on what, and what it did
`run_analytics_query`	✓	Run a custom read-only SQL query over your request logs
`get_uptime_status`	✓	Current origin health per hostname — “is it down right now?”
`list_uptime_incidents`	✓	Outage incident history
`list_recommendations`	✓	NSIN’s findings for the domain and what to change
`get_traffic_usage`	✓	Daily usage split by cached / proxied / bypass / direct
`get_plan_features`	✓	Quota, usage and which features the plan enables

Purging the cache

There are three purge tools, in increasing precision. All of them need cache.edit permission and a plan with cache purge.

purge_cache_by_path is the everyday one — the purge after a deploy:

path: "/assets/*" — everything under a directory
path: "/" — the homepage
path: "/blog/post-1" — one page

Matching rules: * is a wildcard, and a pattern with no * matches as a substring, so /logo.png also matches /img/logo.png?v=2. Anchor the pattern (/assets/*) when that matters. Regular expressions are not supported. Optionally narrow to one hostname or one edge node; omit both and the purge covers every hostname of the domain on every node. Every cache-key variant of a matching URL (query strings, mobile, WebP/AVIF, CORS variants) is purged along with it.

A single call purges up to 10,000 objects and reports truncated: true when the pattern matched more — call it again until that is false. To see what a pattern will hit before running it, pass the same path to list_cache_entries.

purge_cache_entries takes the rows list_cache_entries returned and purges exactly those — use it when a substring pattern would catch more than you meant.

purge_cache drops everything the domain has cached, on every node. It is the blunt option: all traffic goes to the origin until the cache refills.

Both targeted tools also accept mode: "refresh", which drops only the cached copy so the next visitor re-fetches it from the origin, leaving the entry in the listing. If someone is actively deploying, set_developer_mode beats purging in a loop.

Custom analytics queries

run_analytics_query lets an assistant answer analytics questions we don’t ship a dedicated tool for — “top 20 URIs by 5xx count over the last 7 days”, “hourly bandwidth for one subdomain”, “which ASNs sent the most bypass traffic”. You (or your AI) pass a ClickHouse SELECT over the requests table (your request log) and get back columns and rows as JSON (capped at 10,000 rows).

It is strictly sandboxed:

Read-only. Only a single SELECT (or WITH … SELECT) is accepted. Writes, DDL, settings changes, INTO OUTFILE/FORMAT, and table functions (url, file, remote, s3, …) are rejected.
Your data only. Every query is scoped by the engine to the domains your API key can access — there is no way to read another tenant’s rows, even inside a subquery.
Bounded. Execution time and rows read/returned are capped so a heavy query can’t run away.

Useful columns on requests: event_time, domain_id, hostname, method, uri, status, bytesIn, bytesOut, duration (ms), remoteAddr, country, asn, asnOrg, userAgent, cacheStatus (hit/miss/bypass), bypassReason, cacheAge, reqStatus (cache/proxied/direct), isWS, protocol, referer, rayId, node, originStatus, originAddr, error, errFault (which side failed: origin/edge/client), errReason, upstreamDuration (ms origin round-trip), originConnReused, originDialMs, originTlsMs, originTtfbMs, originTlsVersion, ja4, botKind, wafScore, mdConverted, optKind. Bucket time with toStartOfHour(event_time) or toStartOfDay(event_time).

Note that errFault = 'origin' catches failures that never produced a 5xx status — timeouts, refused connections, origins that hung up — so it finds outages a status >= 500 filter misses.

Permissions and safety

The MCP server acts with exactly your account’s permissions. Domain sharing roles apply: with Viewer access to a shared domain, write tools fail on that domain even with a full-access key.
Plan features apply too. Monitoring, request logs, WAF and cache purge are gated on the domain’s plan, and a tool that needs one fails with the reason.
Every MCP action goes through the same validation, audit logging and usage metering as the REST API, and counts toward the key’s rate limit (300 requests/minute).
Revoking the key in the panel disconnects the assistant immediately.