IPGeolocation.io MCP Server

IPGeolocation.io MCP Server

The official IPGeolocation.io MCP server provides 17 tools for accurate IP geolocation, company information, ASN and network data, IP security (VPN, proxy, Tor, bot, hosting and spam detection), abuse contacts, time zones, astronomy, currency data and user-agent parsing.

MCP (Model Context Protocol) connects your client to these tools. Your client or host must support launching a local command over stdio. See Install by Client for configuration examples.

This README describes 2.0.1. The npx commands and Cursor install link below follow npm's latest tag, including future major releases. Restart the server to load an update; npm reuses cached package files when the selected version is unchanged. See the changelog for release details.

Quick Start

1. Create a free IPGeolocation API key.
2. Merge this entry into your client's private configuration, keeping any existing servers. See Install by Client for the configuration location. Cursor also has a one-click installer.

{
  "mcpServers": {
    "ipgeolocation": {
      "command": "npx",
      "args": ["-y", "ipgeolocation-io-mcp@latest"],
      "env": {
        "IPGEOLOCATION_API_KEY": "<YOUR_API_KEY>"
      }
    }
  }
}

3. Reload the server or restart your client.
4. Ask "Where is 8.8.8.8 located?"

Table of Contents

• Quick Start
• Install by Client
• Verify It Works
• Tools by Plan
• Tool Reference
• Prompt Examples
• Example Answers and Tool Output
• Error Codes
• How It Works
• Caching
• Cached Result Pagination
• Environment Variables
• Building from Source
• Docker
• Testing
• Troubleshooting
• Pricing
• Links
• License
• Privacy Policy

Install by Client

Requirements

• Node.js 22 or later for npm/source installs, or a bundle host with a compatible Node.js runtime
• npx available for the npm configuration examples
• An IPGeolocation.io API key for most API tools

Use a supported Node.js release with its latest security updates.

get_my_ip requires no API key. The other 15 API tools require one. get_cached_result reads results already held in memory; use the same API key as the original lookup.

Sign up for a free IPGeolocation API key

Use a private user configuration or your client's secret/environment settings for the real key. Keep keys out of committed project configuration. Each client below uses the Quick Start server entry unless its format differs.

Codex CLI

Set IPGEOLOCATION_API_KEY in your private shell environment before running these commands. The variable reference keeps the key value out of the command you type and its shell history.

codex mcp add ipgeolocation --env "IPGEOLOCATION_API_KEY=$IPGEOLOCATION_API_KEY" -- npx -y ipgeolocation-io-mcp@latest
codex mcp list

Start a new Codex session after adding the server.

Claude Desktop

Install the published server from its Claude Marketplace listing. In Claude Desktop, open Settings > Extensions > Browse extensions, find IPGeolocation.io MCP Server, and configure your API key during installation. See Anthropic's desktop extension guide. The directory distributes the approved published bundle; its refresh can lag GitHub Releases. Check the installed version, or download the versioned bundle below.

For manual npm configuration, open Settings > Developer > Edit Config in Claude Desktop. This opens claude_desktop_config.json:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Merge the Quick Start entry into mcpServers.

Fully quit and reopen Claude Desktop after saving. See the official local-server setup guide.

For clients that install MCP Bundles, download the published v2.0.1 .mcpb bundle from GitHub Releases. The bundle contains the server and its dependencies; its host still needs a Node.js 22+ runtime. Enter your API key when the host prompts for bundle configuration.

Cline

In Cline, open MCP Servers > Configure > Configure MCP Servers. Add the Quick Start entry under mcpServers in the settings JSON. Cline CLI uses ~/.cline/data/settings/cline_mcp_settings.json by default. See Cline's MCP guide.

Restart Cline after saving.

Cursor

One-click install:

Or add to .cursor/mcp.json for this project, or ~/.cursor/mcp.json for all projects. See Cursor's MCP guide.

{
  "mcpServers": {
    "ipgeolocation": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "ipgeolocation-io-mcp@latest"],
      "env": {
        "IPGEOLOCATION_API_KEY": "<YOUR_API_KEY>"
      }
    }
  }
}

Restart Cursor after saving.

Cursor plugin

This repository includes a Cursor plugin for ipgeolocation-io-mcp. The plugin runs locally over stdio and requires Node.js 22 or later.

To install it locally, clone this repository into a new directory under ~/.cursor/plugins/local/, reload Cursor, and open Customize > Plugins.

• Local plugin: set IPGEOLOCATION_API_KEY in the environment that launches Cursor. The plugin reads the key from that environment.
• Marketplace install: set the key in the plugin's private configuration.

With an empty or unset API key, get_my_ip still works. API tools that require a key return a missing-key error without sending a request.

To load a local checkout in Cursor CLI:

agent --plugin-dir /absolute/path/to/ipgeolocation-io-mcp

API requests that need a key use your account's plan and credits. The get_cached_result tool reads saved response pages without making another API request.

Install this server by only one method, because using more than one creates duplicate tool entries. You can keep any other servers in your existing configuration.

Windsurf / Cascade

In Cascade's ... (Actions) menu, choose Open MCP config file in the MCPs section. Merge the Quick Start entry into the file opened by your installed version; configuration paths vary by release. See the vendor's Cascade MCP guide for the active agent and configuration location in newer releases.

Save the file and reload the server in the client.

Kiro Powers

The Kiro Power provides 17 tools for IP geolocation, company information, ASN and network data, IP security (VPN, proxy, Tor, bot, hosting and spam detection), abuse contacts, time zones, astronomy, currency data and user-agent parsing. It runs ipgeolocation-io-mcp locally over stdio, follows npm's latest tag and requires Node.js 22 or later. Import the packaging/kiro folder through Powers > Add Custom Power and follow the setup guide to configure your private API key. Most tools require a key; access and credits depend on your plan.

VS Code / GitHub Copilot

Run MCP: Add Server from the Command Palette, choose a local command, and use npx -y ipgeolocation-io-mcp@latest. Choose Copilot Global for private user settings or .mcp.json for a portable workspace configuration. Add IPGEOLOCATION_API_KEY to the server's environment.

The portable .mcp.json format uses the top-level mcpServers object shown in Quick Start.

Use the server's Start action and approve its configuration when prompted. Existing .vscode/mcp.json files use a top-level servers object instead. See VS Code's MCP setup guide.

Glama

You can try the server on Glama directly. Only IPGEOLOCATION_API_KEY is required. Leave other environment variable fields empty unless you want to change cache, timeout, or output limits.

If you don't have a key yet, create a free IPGeolocation API key.

Any Other MCP Client

Use the ipgeolocation server entry from Quick Start in your client's configuration format. The client must support launching a local stdio process. A client that accepts only a remote MCP URL needs a hosted endpoint; this package supplies the server command.

Verify It Works

Try these after setup:

Tools by Plan

Version 2.0.1 exposes 16 API tools and get_cached_result, a zero-credit page reader. Bulk tools accept 250 inputs by default; the effective ceiling is the lower of the two item-limit settings.

Listed costs apply to successful upstream requests. Cache hits and cached-page reads use no additional credits. The API's X-Credits-Charged response header establishes actual billing; see the credit rules.

Free Plan

1,000 credits per day. These 7 API tools provide their documented core data; get_cached_result also reads previously returned pages for 0 credits:

lookup_company also returns basic ASN ownership on free plans, without company details. lookup_network returns a local error if the API omits the paid network data. The API documentation describes bulk geolocation as paid-only; live checks with a free-plan key also returned basic bulk results. Plan access is determined upstream, so free bulk support should not be assumed from this observation.

Paid Plans

All 17 tools, including the zero-credit cached result reader. Paid plans also add network, company, and extended asn fields to lookup_ip, plus the include parameter for security, abuse, hostname, liveHostname, hostnameFallbackLive, user_agent, geo_accuracy, dma_code, or *.

Credit math for lookup_ip with include:

For current plan details and pricing, see the IPGeolocation pricing page.

Tool Reference

Tool arguments must match the advertised schema. Unknown parameters are rejected locally before any API request, so a misspelled selector cannot silently look up the server's IP.

Coordinates, including conversion selectors, and elevation take string values, such as "24.86" and "10". force_refresh takes a boolean.

lookup_ip

Single IP or domain lookup through the IP Geolocation API. Free and paid. 1 credit for base IP data.

Use this when you need location, timezone, currency, or ASN for one IP address. On paid plans you can add include modules to pull security, abuse, or hostname data in the same call, which avoids extra requests.

Free plan returns base location, country metadata, currency, timezone, and basic ASN. Paid plans add network, company, extended ASN, and the include parameter. Note that domain lookups require a paid plan.

Tip: combining include with fields can cut your credit cost. For example, include=security&fields=security costs 2 credits instead of 3 because you skip the base geolocation response. Similarly, include=abuse&fields=abuse costs 1 credit instead of 2.

Fields enable the optional modules they need. These modules require a paid plan:

Accuracy and DMA values appear inside location. An explicit include keeps its module present even if fields omits it. You can exclude nested fields, but a root named by fields or an explicit include stays present. include=* follows field selection and exclusions without protecting those roots. Cached results follow the same rules as fresh responses.

bulk_lookup_ip

Batch IP lookup. Paid. 1 credit per IP.

Takes an array of IPs or domains (up to 250 by default). The batch limit is the lower of IPGEOLOCATION_MCP_MAX_BULK_ITEMS and IPGEOLOCATION_MCP_MAX_RESULT_ITEMS. Supports the same include, fields, and excludes options as lookup_ip.

get_my_ip

Returns the public IP of the machine running the server. Free. 0 credits. No API key needed.

Takes no parameters. Always hits the network (not cached). Useful as a quick check to confirm the server process is up.

check_security

Threat and anonymity data for one IP through the IP Security API. Paid. 2 credits.

Returns threat score, VPN/proxy/Tor flags, provider names, confidence scores, bot/spam indicators, anonymity flags, and cloud-provider status.

If the same prompt also asks for location or ASN, use lookup_ip with include=security to get them in one request. It costs 3 credits, the same as a separate security check and base lookup. Add abuse to include when abuse contacts are also needed.

bulk_security_check

Batch version of check_security. Paid. 2 credits per IP. Accepts up to 250 inputs by default, subject to the same bulk limits as bulk_lookup_ip.

get_timezone

Current time and timezone details for a location through the Time Zone API. Free and paid. 1 credit.

Accepts IANA timezone names, coordinates, IP addresses, airport codes (IATA/ICAO), or UN/LOCODEs. The response includes timezone offsets, date/datetime variants, current_time, current_time_unix, time_24, time_12, week, month, year, timezone abbreviations, and DST transition details.

With no selector, the API uses the server machine's public egress IP, which can differ from your IP.

Always hits the network (not cached) because it returns the current time.

convert_timezone

Converts a time between two locations. Free and paid. 1 credit.

Provide one source selector and one destination selector using an IANA time zone, coordinates, a city/address, an airport code, or a UN/LOCODE. If you omit time, the API converts the current time. This tool does not accept IP selectors or lang and always calls the API (not cached).

get_astronomy

Sun and moon data for one location on one date through the Astronomy API. Free and paid. 1 credit.

Returns sunrise, sunset, moonrise, moonset, morning and evening twilight, solar noon, day length, moon phase, sun/moon status flags, and live sun/moon position (altitude, azimuth).

This tool always calls the API, including when date is set; it does not use the response cache.

get_astronomy_time_series

Astronomy data for a date range, up to 90 days. Free and paid. 1 credit per request.

Each daily entry includes mid_night, night_end, morning, sunrise, sunset, evening, night_begin, sun_status, solar_noon, day_length, moon_phase, moonrise, moonset, and moon_status. Use this instead of calling get_astronomy repeatedly for a range.

parse_user_agent

Parses one UA string into browser, device, OS, and engine data through the User-Agent API. Paid. 1 credit.

Also classifies bots and crawlers. Note: this parses the uaString you pass in. It does not infer a caller UA from the MCP connection itself.

Returns name, type, version, device, engine, and operating_system.

bulk_parse_user_agent

Batch version of parse_user_agent. Paid. 1 credit per string.

Takes up to 250 strings per request by default. The batch limit is the lower of IPGEOLOCATION_MCP_MAX_BULK_ITEMS and IPGEOLOCATION_MCP_MAX_RESULT_ITEMS.

lookup_company

Company details and ASN ownership for one IP. Company details require paid; free plans return basic ASN only. 1 credit.

Returns company and asn objects. lookup_ip returns the same data plus location, timezone, and more. Use this when the company/ASN pair is all you need and you want a smaller response.

lookup_currency

Currency, country calling code, TLD, and languages for one IP. Free and paid. 1 credit.

Returns currency and country_metadata objects.

lookup_network

Route prefix, connection type, and anycast status for one IP. Paid. 1 credit.

Returns a network object with connection_type, route, and is_anycast. If the API omits network data, the tool returns a local error instead of an empty success. This does not invent an upstream HTTP status, and the API call may already have consumed credits.

lookup_asn

Network allocation and relationship data through the ASN API. Paid. 1 credit.

Use this tool for ASN details and optional peers, upstreams, downstreams, routes, and WHOIS datasets. Request the datasets you need with include, then filter the cached result to retrieve different slices. lookup_ip also provides ASN metadata as part of an IP lookup.

fields automatically enables optional datasets it references, such as upstreams.as_number. excludes alone does not request extra datasets.

get_abuse_contact

Abuse contact details for one IP through the IP Abuse Contact API. Paid. 1 credit.

Returns the abuse route, country, contact name, organization, address, email addresses, and phone numbers. If you also need geolocation or security data for the same IP, use lookup_ip with include=abuse (or include=security,abuse) to get everything in one call.

get_cached_result

Read a page from a previous oversized result. Free and paid. 0 credits. Added in 2.0.1.

Use the same server process and API key. Page reads extend the snapshot's idle expiry and never call the API, including on failure. See Cached Result Pagination for the response fields and a complete recovery example.

Prompt Examples

For a paid IP investigation, call lookup_ip once with ip: "49.12.212.42", include: "security,abuse", and fields: "location,company,asn,security,abuse". This combines location, ownership, threat indicators, and abuse contacts for an estimated 4 credits. Describe the evidence and any unknown fields; apply the user's policy before recommending an allow/block decision.

For a mixed batch, call bulk_lookup_ip with ips: ["8.8.8.8", "10.0.0.1", "1.1.1.1"] and fields: "location.country_name". Inspect structuredContent.errors to identify the private input and use the successful records directly. The two valid IP lookups cost an estimated 2 credits; the private IP is not billed. Split larger batches at the advertised input limit.

An ASN identifies a network that announces IP routes. Anycast lets one IP be served from multiple locations, so its location need not identify the server that handled a particular request. A bogon address is reserved or otherwise not publicly routable. Upstream in an error means the IPGeolocation.io API contacted by this server.

IP location is an estimate. Use include=geo_accuracy where available to assess uncertainty. Threat scores and VPN/proxy flags are signals for investigation, not guarantees of harmless or malicious traffic. Treat missing, null, or unknown indicators as unavailable, not as false. Combine location, ownership, security signals, and abuse contacts with your own allow/block policy.

Inspect IP threat indicators

• Show the threat indicators and city for 49.12.212.42 so I can assess it against our network policy.
• Check these IPs for VPN, proxy, Tor, bot, and spam indicators: 49.12.212.42, 2.56.12.11, 8.8.8.8
• For 49.12.212.42, tell me the threat score, whether it is a cloud provider, and whether it looks like a relay.

Find who owns an IP

• Who uses 1.1.1.1 and which ASN routes it?
• For AS24940, list upstream ASN numbers only.
• Is this IP anycast and what route prefix is announced for it: 1.1.1.1

Get abuse contacts

• For IP 2.56.12.11, give me the abuse contact email, phone number, and organization.
• I need the abuse contact for 1.0.0.0 and the network route involved.
• For this IP, show me the abuse contact details only: 1.1.1.1

Timezone lookups and conversions

• What time is it in Tokyo right now?
• Convert 2026-03-07 09:30 from New York to Tokyo time.
• What is the current local time at JFK airport?

Sunrise, sunset, and moon data

• Give sunrise and sunset for London on 2026-06-21.
• Show sunrise times in Karachi from 2026-03-10 to 2026-03-15.
• For New York, give me moon phase and day length on 2026-07-17.

Parse user-agent strings

• Parse this user agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_11_2) AppleWebKit/601.3.9 (KHTML, like Gecko) Version/9.0.2 Safari/601.3.9
• Parse these user agents in bulk and tell me the browser, OS, and device type for each.
• Does this user agent look like a crawler or bot? Mozilla/5.0 (Linux; Android 6.0.1; Nexus 5X Build/MMB29P) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/41.0.2272.96 Mobile Safari/537.36 (compatible; Googlebot/2.1; +http://www.google.com/bot.html)

Example Answers and Tool Output

The text answers below are illustrative; exact wording depends on the model. JSON blocks are shortened examples, not current measurements. Location, ownership, contacts, and current-time fields can change.

Single IP lookup

Prompt: Locate 91.128.103.196 and give me the country, city, ASN, and local time.

Example answer: 91.128.103.196 is in Stockholm, Sweden. ASN is AS1257, operated by Tele2 Sverige AB. Timezone is Europe/Stockholm, local time was 2026-02-12 18:36:54.

{
  "ip": "91.128.103.196",
  "location": {
    "country_name": "Sweden",
    "state_prov": "Stockholms lan",
    "city": "Stockholm"
  },
  "asn": {
    "as_number": "AS1257",
    "organization": "Tele2 Sverige AB",
    "country": "SE"
  },
  "time_zone": {
    "name": "Europe/Stockholm",
    "current_time": "2026-02-12 18:36:54.401+0100"
  }
}

Time conversion

Prompt: Convert 2025-01-21 13:42:52 from DXB to LHR.

Example answer: 2025-01-21 13:42:52 in Dubai converts to 2025-01-21 09:42:52 in London. The difference is 4 hours.

{
  "original_time": "2025-01-21 13:42:52",
  "converted_time": "2025-01-21 09:42:52",
  "diff_hour": 4,
  "diff_min": 240
}

Abuse contact

Prompt: Give me the abuse contact for 1.0.0.0.

Example answer: The abuse contact for 1.0.0.0 is IRT-APNICRANDNET-AU in Australia, covering route 1.0.0.0/24. Email: [email protected].

{
  "ip": "1.0.0.0",
  "abuse": {
    "route": "1.0.0.0/24",
    "country": "AU",
    "name": "IRT-APNICRANDNET-AU",
    "address": "PO Box 3646, South Brisbane, QLD 4101, Australia",
    "emails": ["[email protected]"]
  }
}

Error Codes

Tool failures return isError: true, readable text, and structuredContent.error with a redacted, length-limited message, category, retryability, and guidance:

Input validation uses the registered Zod schemas and the same structured error envelope as execution failures. Unknown tools and malformed MCP requests produce protocol errors. Retry guidance does not guarantee that a retry will succeed or be free: a local timeout can happen after the API processed a paid request. HTTP 429 is a credit/subscription limit, so immediate automatic retries are disabled; retry after credits become available.

The following table applies to HTTP errors actually received from the API:

Exact status codes can vary by endpoint and request mode. If an upstream endpoint returns a status outside this table, the server does not guess the cause. It passes the upstream status and message through with category: "undocumented_api_error" and adds guidance for the MCP client to explain the response as an undocumented upstream status without inventing a cause.

Bulk text preserves the API's per-item format. Structured results contain items in input order plus an errors list. Each failed item includes its zero-based input_index, original input, source, category, message, retryability, and guidance. An item status is present only if supplied by the API; message-only errors do not acquire an invented HTTP status. A mismatched result count returns a whole-call error rather than assigning unreliable input identifiers. Inspect every record; overall request success does not establish that every item succeeded, and successful items should not be retried.

How It Works

Your MCP client uses this server to call the IPGeolocation.io v3 APIs.

At runtime:

1. Your MCP client starts the server process.
2. The client reads the tool list.
3. When a prompt matches a tool, the client calls it.
4. The server validates arguments and returns a cached result or calls the API. It returns text content plus structuredContent; all 17 tools advertise an output schema.

Normal text responses keep the API's JSON format, or a plain IP string for get_my_ip. Oversized successful responses contain a page in both text and structured content.

Authenticated upstream requests send the API key in the x-ipgeolocation-api-key header. The key is not added to request URLs.

Server instructions and tool descriptions explain selection, costs, limits, and interpretation to the MCP client. Clients decide how to expose this guidance to their assistant. Returned WHOIS text, user agents, hostnames, and error messages are untrusted data and must not override those instructions. An omitted IP selector refers to the server machine's public egress IP, which can differ from the end user's IP. Security tools query IP intelligence; they do not invoke the separate browser-based live VPN/proxy detection API.

lookup_company, lookup_currency, and lookup_network are wrappers around parts of the full IP lookup response. They exist as separate tools so MCP clients can discover them when a user only needs one piece of data.

Caching

The server stores cached responses in process memory and shares identical requests already running for the same API key. Sessions using the same key share the cache namespace; different keys use separate namespaces.

Both use IPGEOLOCATION_MCP_CACHE_TTL_MS (300000 ms by default). The cache holds up to 500 entries by default, shared by normal responses and page snapshots. This limits entry count, not total bytes or memory. A complete snapshot can be much larger than an individual page.

Expiry, eviction, or process restart can make cached data unavailable. get_cached_result then fails locally without refetching. Repeating the original lookup after a cache miss can consume credits again.

force_refresh: true starts a new upstream request, even if an identical request is running. Older requests cannot overwrite its cached result.

Cached: lookup_ip, bulk_lookup_ip, check_security, bulk_security_check, lookup_company, lookup_currency, lookup_network, parse_user_agent, bulk_parse_user_agent, lookup_asn, get_abuse_contact, get_astronomy_time_series

Always live (not cached): get_my_ip, get_timezone, convert_timezone, get_astronomy

Environment Variables

Bulk batches above the effective item limit are rejected before an upstream request, without using API credits. Raising only IPGEOLOCATION_MCP_MAX_BULK_ITEMS does not raise the effective limit; raise both item settings if larger batches are needed.

The response limit includes both text and structured content, JSON escaping, and UTF-8 encoding. Successful results retain every record and nested array through cached pages when needed. Oversized errors shorten only their messages, preserving their category, retryability, guidance, and any status.

Cached Result Pagination

get_cached_result costs 0 credits and never calls the API, including when its cursor is invalid, expired, evicted, or unavailable after a server restart. It accepts one required cursor, returned in page.cursor or page.next_cursor. Use the same server process and API key.

A large original lookup returns { "page": ... } instead of its normal data or items/errors. Each page contains:

Read each next_cursor with get_cached_result, concatenate the json_fragment strings in offset order, and parse the joined JSON once. A fragment is not necessarily valid JSON on its own. This also handles a single record or string larger than one page. Bulk results reconstruct the complete { "items": [...], "errors": [...] }, preserving input identifiers and statuses. Single lookups reconstruct { "data": ... }.

let response = await client.callTool({ name: "bulk_lookup_ip", arguments: { ips } });
if (response.isError) throw new Error(response.content[0].text);
let complete = response.structuredContent;
if (complete.page) {
  let json = "";
  for (;;) {
    const page = response.structuredContent.page;
    json += page.json_fragment;
    if (!page.next_cursor) break;
    response = await client.callTool({
      name: "get_cached_result", arguments: { cursor: page.next_cursor }
    });
    if (response.isError) throw new Error(response.content[0].text);
  }
  complete = JSON.parse(json);
}

Use cursors to retrieve remaining data. The cache rules explain snapshot expiry, eviction, API-key isolation, and when a new lookup can consume credits.

Building from Source

git clone https://github.com/IPGeolocation/ipgeolocation-io-mcp.git
cd ipgeolocation-io-mcp
npm ci --ignore-scripts
npm run build

Run it directly:

node dist/cli.js

Set IPGEOLOCATION_API_KEY through your shell's private environment settings before running the command. The process waits for MCP messages on stdin; it does not display an interactive prompt.

Inspect with MCP Inspector:

npx @modelcontextprotocol/inspector node dist/cli.js

Docker

docker build -t ipgeolocation-mcp .
docker run --rm -i -e IPGEOLOCATION_API_KEY ipgeolocation-mcp

Set IPGEOLOCATION_API_KEY in your private shell environment first. The image includes Node.js and runs as the unprivileged node user. Docker's -i flag keeps stdin open for MCP messages. Omit -t so the protocol uses ordinary stdin/stdout.

The final image runs the compiled server directly. npm and Yarn are available in the build stage; the runtime image contains Node.js and the installed production dependencies.

After building the image, configure a local-command MCP client with:

{
  "mcpServers": {
    "ipgeolocation": {
      "command": "docker",
      "args": ["run", "--rm", "-i", "-e", "IPGEOLOCATION_API_KEY", "ipgeolocation-mcp"],
      "env": {
        "IPGEOLOCATION_API_KEY": "<YOUR_API_KEY>"
      }
    }
  }
}

Testing

Client setup examples document configuration formats. Record GUI and operating-system installation checks separately from protocol, package, and Docker tests; see the release checklist.

npm test               # full suite
npm run test:unit      # unit tests only
npm run test:integration  # integration tests only

Live smoke tests use a real stdio MCP client and spend API credits. IPGEOLOCATION_FREE_KEY with npm run test:live:free checks the seven core tools, paid-feature restrictions, ownership/network fallbacks, and the observed bulk entitlement. It reports when bulk access differs from the documented paid-only policy. IPGEOLOCATION_PAID_KEY with npm run test:live:paid checks all 16 API tools, paid modules, structured validation errors, mixed valid/private bulk inputs, and the local cached-result reader. Testing a paid key does not verify free-plan entitlement restrictions.

After building, verify complete cached pagination with either key:

node tests/pagination.stdio.mjs free  # uses IPGEOLOCATION_FREE_KEY
node tests/pagination.stdio.mjs paid  # uses IPGEOLOCATION_PAID_KEY

Each run makes one six-IP API request and checks the reported six credits. It compares the recovered records against a hash of the complete upstream response and verifies that page reads, replay, expiry, and process restart make no additional API calls. The test temporarily lowers the child server's response budget and cache TTL; your server configuration stays unchanged.

To check the same behavior in the image built above:

node tests/pagination.stdio.mjs free docker run --rm -i \
  -e IPGEOLOCATION_API_KEY -e IPGEOLOCATION_MCP_MAX_RESPONSE_CHARS \
  -e IPGEOLOCATION_MCP_CACHE_TTL_MS -e NODE_OPTIONS ipgeolocation-mcp

Troubleshooting

Client uses an old tool after updating: Restart the server process and check its version in your client's MCP connection details. Compare it with the latest npm version. Rebuild source and Docker installations before restarting them.

API key not configured: Set IPGEOLOCATION_API_KEY in your client's server configuration. This local error has no HTTP status and makes no upstream request.

401 errors: The API rejected the key or requested capability. Check that the key is active and the plan supports the tool. Domain lookups in lookup_ip also require a paid plan.

423 errors: Use a public routable IP. Private/bogon addresses such as 10.0.0.1 and reserved documentation addresses such as 203.0.113.42 cannot be used as successful geolocation examples.

Local 504 timeouts: With source: "mcp_server", the request or response-body read exceeded the deadline. Check the connection or increase IPGEOLOCATION_REQUEST_TIMEOUT_MS (default: 15000 ms, max: 120000 ms). With source: "ipgeolocation_api", 504 is an HTTP response from the API.

Redirect blocked: The server received an unexpected HTTP redirect and stopped before forwarding credentials or request data. Do not retry automatically. Investigate the API endpoint or contact support with the sanitized error.

Cached result unavailable: The cursor belongs to another key or process, or the snapshot expired or was evicted. The failed page read made no API call. A new lookup can consume credits.

Client asks for an MCP URL: This package supplies a local stdio command. Use a client or host that can launch it; a remote-only connector needs a hosted MCP endpoint.

Pricing

For current plan details, credits, and pricing, see the IPGeolocation pricing page.

Links

• IPGeolocation Website
• IPGeolocation API Documentation
• IPGeolocation Pricing
• Create a Free IPGeolocation API Key
• Changelog
• Security Policy
• Contributing
• Release Checklist

License

MIT

Privacy Policy

API lookups send the requested inputs to IPGeolocation.io, which also receives the server machine's public egress IP. The server caches results in process memory. Your MCP client or hosting provider controls its own configuration and log storage. See the Security Policy for retention details and the IPGeolocation Privacy Policy for provider handling.