DNSFilter

DNSFilter MCP Server

MCP server for DNSFilter's DNS security/filtering API - organizations, networks (sites), filtering policies, content/application categories, custom block pages, and the full traffic/threat reporting surface, for AI assistants and the WYRE Conduit gateway.

Scope

This is a deliberately narrow, read-only v1 surface, hard-scoped to organizations, networks, policies, categories, block pages, and reporting. DNSFilter's REST API (api.dnsfilter.com) has 217 documented paths across a much broader surface - MSP/distributor management, billing, users, roaming-client agents, enterprise SSO connections, API key management, scheduled reports, and more. None of that is implemented here, by design, not by oversight. Every tool in this connector maps 1:1 to a real, documented GET operation in DNSFilter's own published OpenAPI spec (https://api.dnsfilter.com/docs.json) - there is no write, update, or delete tool anywhere in this codebase.

Hard-excluded (every write/mutation operation on every in-scope resource) - never implemented:

• POST/PATCH/DELETE /v1/organizations* (create, bulk_update, promote_to_msp, update, cancel, destroy) - organization mutations.
• POST/PATCH/DELETE /v1/networks* (create, bulk_create/update/destroy, update, destroy, secret_key rotate/revoke/generate) - network mutations, including LAN IP and subnet writes.
• POST/PATCH/DELETE /v1/policies* (create, update, destroy, application_update, add/remove_allowed_application, add/remove_blacklist_category, add/remove_blacklist_domain, add/remove_blocked_application, add/remove_whitelist_domain, permissive_mode update) and all of /v1/policies/bulk/* (bulk allow/block-list add/remove) and /v1/policy_ips writes - every policy-content mutation.
• POST/DELETE /v1/block_pages* (create, destroy) and PATCH /v1/block_pages/{id} (update) - block-page mutations.
• Every write endpoint on resources this connector doesn't otherwise touch at all: API keys (create/destroy/revoke), users and organization users (create/update/destroy/resend_invite), billing/invoices, distributors (MSP/sub-org provisioning, SKUs, cancellation), enterprise SSO connections, scheduled policies/reports, roaming-client user-agents (bulk delete/update/cleanup, dequeue-uninstall), domain notes, IP/MAC address management, and trials.

They can be added as a follow-up if there's demand, after a deliberate scope decision - not by default.

Deferred (read-only, not implemented in this v1 - not a security exclusion): a handful of GET-only sub-resources and dashboard-adjacent endpoints exist under the same resource groups this connector covers, but aren't wired up yet: network subnets (GET /v1/networks/{id}/subnets*), network LAN IPs (GET /v1/networks/{id}/lan_ips*), policy IPs (GET /v1/policy_ips*), the unpaginated */all variants of every list endpoint (redundant with the paginated list tool), and administrative */counts endpoints. These can be added later without any scope-policy change, since they're already read-only.

Authentication

DNSFilter authenticates with a static API key, generated in the DNSFilter dashboard under Account Settings -> Security -> API Keys (an account supports up to 5 active keys at a time; see DNSFilter's own API Keys article). Unlike most sibling WYRE Conduit connectors, the key is sent as the raw header value - Authorization: , with no `Bearer` prefix - per DNSFilter's own API docs: *"Authentication is done by setting the Authorization request header. The header value is the API key itself. For example: Authorization: eyJ..."*.

Credential scope: two separate claims, at deliberately different confidence levels

• Structurally verified (checked directly, stated with full confidence): this connector's own code makes zero mutating calls - every function in client.ts is a GET, and no POST/PUT/PATCH/DELETE call exists anywhere in src/ (enforced by tool-scope.test.ts's forbidden-token check). Also structurally verified, directly against DNSFilter's own published OpenAPI spec (api.dnsfilter.com/docs.json, fetched directly): an API key is generated per user account (the ApiKey resource schema carries a user_id field) and its own creation request (ApiKeyCreateSpec) accepts only name and expiry - there is no scope/role/permission field chosen at key-creation time. A key's effective capability is therefore inherited from whatever role the DNSFilter user account it was generated under holds, not selected independently when the key itself is created. That same spec's Membership/OrganizationUser role enums explicitly include a "read_only" value alongside "administrator" (and, on the MSP/distributor membership shape, alongside "network_administrator"/"network_support"/"support") - so a genuinely restricted DNSFilter user role is a first-class, documented concept in the API itself, not just dashboard-UI copy.
• Vendor-documented, not independently verified (hedged deliberately): DNSFilter's own help center separately describes a dashboard-level Read Only user role ("Users with more restrictive roles like Read Only... have hidden dashboard tabs and panels"). Combined with the structural finding above, generating an API key under a dedicated Read-Only-role user account *should* yield a key whose calls DNSFilter's API server rejects for any write endpoint - but whether the API actually enforces that role server-side against write calls (versus the restriction being dashboard-UI-only) has not been tested by WYRE. Nobody sent a live write call against a Read-Only-role key to confirm, correctly: that would be a mutating test against a real account, not something to run without consent. Do not read this README as having established that a Read-Only-role API key cannot perform writes against DNSFilter's API - only that (a) DNSFilter's spec models a read_only role as a real, first-class concept, and (b) this connector's own code never attempts a write regardless of which key it's given.

In gateway mode the key arrives per-request via the X-DNSFilter-Api-Key header; in local/stdio mode it's read once from DNSFILTER_API_KEY. The stdio transport does not use CONDUIT_S2S_SECRET.

HTTP authentication

MCP_TRANSPORT=http refuses to start when CONDUIT_S2S_SECRET is empty or a known placeholder (changeme, change-me, placeholder, secret, example, replace-me, replace-with-a-real-secret, your-s2s-secret, your-secret, s2s-secret). The process logs an error and exits non-zero. It never prints the secret. A placeholder is refused even if MCP_ALLOW_INSECURE_DEV=1 is set.

MCP_ALLOW_INSECURE_DEV=1 is local development only. It starts HTTP without a secret, logs a warning, does not check X-Gateway-S2S, and listens on 127.0.0.1 only. A non-loopback MCP_HTTP_HOST (including 0.0.0.0) is ignored in that mode and the override is logged.

When a real secret is set, MCP_HTTP_HOST is honored. Unset, the bind address is 127.0.0.1. The container image sets MCP_HTTP_HOST=0.0.0.0 and AUTH_MODE=gateway.

Every /mcp request must then include X-Gateway-S2S:

t=<unix seconds>,v1=<hex hmac-sha256>

v1 is HMAC-SHA256 of the ASCII string t= , keyed with CONDUIT_S2S_SECRET, hex-encoded. The timestamp must be within 300 seconds of the server clock. A missing or invalid header returns 401.

In AUTH_MODE=gateway, /mcp also requires X-DNSFilter-Api-Key. A request missing that header returns 401 and does not fall back to DNSFILTER_API_KEY. /health stays unauthenticated and does not read credentials.

Request limits

• HTTP request bodies are capped at 256 KiB (262144 bytes). After the X-Gateway-S2S check (and, in gateway mode, the X-DNSFilter-Api-Key presence check) passes, a Content-Length above the cap returns 413 with a JSON-RPC error; bodies without Content-Length are capped by the MCP SDK transport. Unauthenticated requests still get 401 first.
• tools/call arguments are capped at 10,000 elements (every array item and object key at any depth). Larger inputs return an isError tool result with a fixed message (no caller input echoed) and are never dispatched.

These limits address CVE-2026-111768 / GHSA-46jr-4wx8-v453 (memory exhaustion during validation) and GHSA-f897-62hf-6gxq (no request body / batch cap).

Configuration

Copy .env.example and set the variables below.

Tools

57 read-only tools, one per documented DNSFilter GET operation across six resource groups.

Organizations

• dnsfilter_list_organizations - list organizations visible to this API key.
• dnsfilter_get_organization - get one organization's basic information.
• dnsfilter_get_organization_settings - get organization-level settings.

Networks

• dnsfilter_list_networks - list networks (sites/locations).
• dnsfilter_get_network - get one network's basic information.
• dnsfilter_lookup_network_by_ip - find the network associated with a source IP address.
• dnsfilter_get_networks_geo - get every network's geo/location metadata only.

Policies

• dnsfilter_list_policies - list filtering policies.
• dnsfilter_get_policy - get one policy's full configuration (allow/block lists, categories, applications).
• dnsfilter_get_application_policies - get which policies allow/block a specific application.
• dnsfilter_get_policy_permissive_mode - get whether a policy has permissive (monitor-only) mode enabled.

Categories

• dnsfilter_list_categories - list content-filtering categories.
• dnsfilter_get_category - get one content category's details and hierarchy.
• dnsfilter_list_application_categories - list application categories.
• dnsfilter_get_application_category - get one application category's details.

Block Pages

• dnsfilter_list_block_pages - list custom block pages.
• dnsfilter_get_block_page - get one custom block page's configuration.

Metrics

• dnsfilter_get_organization_usage - get organization usage (plan, users, total requests) over a date range.
• dnsfilter_get_organization_usage_detailed - get the detailed per-metric breakdown of organization usage.

Traffic Reports

DNSFilter's entire reporting surface is GET-only, so all 38 documented report operations are implemented with no curation:

• dnsfilter_get_qps, dnsfilter_get_qps_active_agents, dnsfilter_get_qps_active_organizations - queries-per-second time series.
• dnsfilter_get_query_logs - raw DNS query log entries (domain, requester, category, allow/block result). The mos