APIs and data sources an agent can call or load, from AI model APIs to Australian government, property and YouTube data.
38 entries. Latest source check .
What a source check covers
Documentation checked means the source was read. It does not mean the software was installed, tested or security audited.
Reading an entry does not authorise installation, credential access or spending. Tech Horizon Labs projects are labelled, and there are no affiliate links or paid placement.
Agents and apps that verify or pre-fill Australian business details.
Access and use
Free. Requires registering and accepting the web services agreement to receive an authentication GUID by email; keep the GUID out of code and chats. Its JSON service covers ABN, ACN and name searches.
Domain Group's API packages for Australian property data: address suggestions, agents and listings, listing management, price and rental estimates, auction results, suburb statistics and schools.
Agencies, property apps, brokers and fintechs building on Australian real estate data.
Access and use
Sign up with a corporate email, build in the sandbox, then go live per package. Pricing and limits depend on the package, so check the plan before building. Listing Management can create and update live listings.
Agents and apps that check or pre-fill UK company details.
Access and use
Free. Needs a Companies House account and an API key sent by HTTP Basic auth; keep the key out of code and chats. Up to 600 requests in 5 minutes. Officer and PSC records are about people.
Agents and analysts researching US public companies and their filings.
Access and use
No key. The SEC asks automated requests to declare a name and contact email in the User-Agent header and to stay under 10 requests a second. Covers SEC filers, mostly public companies.
MBIE's API for searching the New Zealand Business Number Register and reading business details, including companies from the Companies Office registers.
Agents and apps that check or pre-fill New Zealand business details.
Access and use
No fee. Needs a signed MBIE API Access Agreement, an approved subscription and a subscription key sent in a header. Passed-on data must carry the search time and "Data sourced from the [Register name]".
Agents and scripts that deploy or manage sites and Workers on Cloudflare.
Access and use
Use a permission-scoped API token rather than the legacy global key, and give it only the zones and products the job needs. The published limit is 1,200 requests per five minutes per user or account token, and 200 per second per IP, with extra limits on some endpoints. Free plans exist, but metered products bill separately. Permission to manage an account is not permission to share its customers' data.
Authenticate with a Console API key or workload identity federation. Billing is per token, and some Max, Ultra and Team subscriptions include monthly API credits that have to be claimed. Rate limits depend on the organisation's usage tier and the model class. Commercial API inputs and outputs aren't used for training by default, with opt-in and feedback exceptions.
Authenticate with an API key or a short-lived workload identity token; admin keys cover a separate administrative surface. Billing is per use, and API billing is separate from a ChatGPT plan. Limits are set per organisation and project and vary by model. API data isn't used for training by default, but abuse-monitoring logs can be kept for up to 30 days unless approved retention controls apply.
Keys belong to a Google Cloud project, and limits (requests per minute, tokens per minute and requests per day) apply per project, with daily limits resetting on Pacific time. There is an unpaid tier for eligible models. Under the unpaid terms, Google may use what you send to improve its products and people may review it, so don't send confidential or client data on that tier. Paid use has different terms. In the European Economic Area, Switzerland and the UK, the paid data terms apply to unpaid use too. These terms took effect on 23 March 2026.
Agents and dashboards that report on a site's Google search traffic.
Access and use
Private site data needs OAuth from a user with access to that property; request the read-only scope when you only read. Search Analytics allows 1,200 queries per minute per site and user, and URL inspection allows 2,000 a day per site. The data belongs to the property, so access follows whoever the owner has granted.
Agents and apps that look up local businesses, addresses or opening hours.
Access and use
For production, needs a Google Cloud project with billing and a restricted API key. For prototyping, the Maps Demo Key needs no billing account or credit card; it has a daily limit per API and covers a set list of APIs, including Text Search, Nearby Search and Autocomplete. Pricing is per SKU, and the fields you request can move a call to a dearer SKU, so ask only for the fields you need. Storing or caching results is restricted, except place IDs, and results come with display and attribution rules.
Agents that work inside a business's Microsoft 365 tenant.
Access and use
Uses Microsoft identity platform tokens with delegated or application permissions, and some permissions need an admin to consent for the tenant. Many calls come with the user's licence, but some APIs are metered. Throttling varies by service, so honour 429 responses and their Retry-After header. A token reaches only the tenant and data its permissions allow.
Teams that want to try or switch between models from several providers through one key.
Access and use
Authenticate with an API key, or use the documented OAuth flow for users. Some models have free variants, limited to 20 requests a minute with a daily ceiling that depends on credits bought; paid models use credits. Each provider has its own data policy. OpenRouter offers privacy settings and zero-retention routing, but legal and safety exceptions still apply.
Agents and SEO tools that need search and keyword data on demand.
Access and use
Uses HTTP Basic authentication with your API login and a generated API password, which differs from the dashboard password. Pay as you go, with trial credit for new accounts and a sandbox of sample responses. The default limit rose to 10,000 requests a minute, though some endpoints have their own limits. Label which data is measured and which is modelled, and check the terms before passing data on.
Use a restricted API key with only the permissions the job needs, and test in a sandbox before using live keys. A restricted key for an autonomous agent can be tagged as an agent key; agent-tagged keys need a designated reviewer to approve sensitive actions, and from December 2026 Stripe's MCP server stops accepting full-access secret keys and restricted keys without the Agent tag. The published limits are 100 requests a second in live mode and 25 in a sandbox, with a 25-a-second limit on most endpoints and further exceptions. Read requests must also average no more than 500 per transaction over a rolling 30 days, with a minimum allowance of 10,000 a month. A key that can read invoices shouldn't be able to refund or pay out, so scope it accordingly.
Agents and apps that send email from a verified domain.
Access and use
Authenticate with an API key that can be limited by permission and domain. The free plan covers 3,000 emails a month, up to 100 a day. The default API limit is 10 requests a second per team, separate from sending quotas. Recipients' details and email content are customer data, and API access doesn't give consent to email anyone.
Agents that read or update a team's Notion workspace.
Access and use
Connect with an internal connection token, a personal access token or OAuth; each reaches only the pages shared with it. A personal token acts as its user. Limits are 600 requests a minute per connection on Business and Enterprise plans and 180 on other plans, plus workspace limits, so honour Retry-After. Querying a data source has its own limit of about 3 requests a second per connection.
Teams producing video with AI avatars or translated voice.
Access and use
The API uses an API key with scopes and bills against an API plan. The MCP server and CLI sign in as your web account and use its subscription credits instead, so a web plan isn't API credit. Limits vary by endpoint and plan. Using a person's likeness or voice needs their consent.
Developers trying open models before choosing where to run them.
Access and use
Needs an NVIDIA developer account and API key. Developer Program access is for prototyping, research, development and testing only; production, which NVIDIA defines to include anything serving real end users, needs an NVIDIA AI Enterprise licence. The API trial terms bar confidential information and, unless a service allows it, personal data, and say NVIDIA sets use limits; the docs give no rate limit figure. The model list at https://integrate.api.nvidia.com/v1/models answers without a key, and a retired model's id returns HTTP 410 with its end-of-life date. Each model's licence and terms apply separately.
Agents that read or manage code and project work on GitHub.
Access and use
Use a fine-grained personal access token, a GitHub App or OAuth; workflows get a built-in token. Unauthenticated calls to public endpoints allow 60 an hour, and an authenticated user gets 5,000 an hour, with separate limits for search and apps. A public repository still has its own licence, so check it before reusing code.
The Management API uses a scoped personal access token or OAuth. Each project's data APIs use their own keys and row-level security, which is a separate layer. The free plan covers two active projects, and inactive ones pause. Never expose a management or service key in a client app.
Google's official API for targeted queries on a channel's own analytics, such as views, ratings, audience age, gender and country, and estimated revenue.
Channel owners, agencies and the agents they authorise to report on a channel's performance.
Access and use
Needs OAuth consent from the channel or content owner, so it cannot read another creator's private analytics unless they grant access. The yt-analytics.readonly scope covers activity metrics, revenue needs the separate yt-analytics-monetary.readonly scope, and Google sets a quota cost for each query. The YouTube API Services Terms and Developer Policies apply, including limits on storing data and on combining data from different content owners.
Analysts who want a channel's full daily history in their own database.
Access and use
Needs OAuth consent from the channel or content owner, with the same read-only and monetary scopes as the Analytics API. A new reporting job backfills 30 days, the first files can take 48 hours, and reports stay downloadable for 60 days, or 30 days for backfilled history. The YouTube API Services Terms and Developer Policies limit how long stored data can be kept without a refresh.
Developers who want transcripts through a paid API instead of running their own scraper.
Access and use
Needs a Supadata account and API key, or OAuth sign-in for the hosted MCP server at api.supadata.ai/mcp; the local server reads SUPADATA_API_KEY. The free plan gives 100 credits a month at one request a second, a transcript costs 1 credit, and a generated transcript costs 2 credits a minute. The MCP server code is MIT licensed.
The public API and database behind SponsorBlock, with crowdsourced timestamps for sponsor reads, self-promotion, intros and other segments in YouTube videos.
Analysts measuring how often and how long channels run sponsor segments.
Access and use
Reading segments needs no account or key, and the full database can be downloaded. The API and database are under CC BY-NC-SA 4.0 unless the maintainer gives permission, so commercial use needs that permission and reuse needs attribution. Automated submissions are not allowed, and segments are community votes rather than verified facts.
A public API that returns like counts and estimated dislike counts for YouTube videos, built from archived dislike data and estimates extrapolated from its extension's users.
Analysts who want a rough measure of negative reaction to a video.
Access and use
No account or key is needed, and each client is limited to 100 requests a minute and 10,000 a day. Third-party use must credit the project with a link to its repository or returnyoutubedislike.com. Dislike figures for the period since YouTube hid them are estimates, so label them as such.
Agents that need to see what ranks in YouTube search for a keyword.
Access and use
Needs a SerpApi account and private API key; the free plan covers 250 searches a month at 50 an hour, and the Starter plan is US$25 a month for 1,000. Cached searches are free, and an output option returns Markdown for agents. SerpApi's U.S. Legal Shield, its cover for scraping and parsing search data, starts on the Production plan.
Agencies building creator search or vetting into their own tools.
Access and use
A free trial gives 80 one-time credits, and ongoing access uses a Nox Key set up through an account manager with quoted pricing; each key is limited to 60 requests a minute. The site's terms forbid crawling NoxInfluencer data, so use the API rather than automating the web app. Contact details it returns are personal information.
Modash's Discovery and Raw APIs for searching and analysing YouTube, Instagram and TikTok creators, with audience demographics, lookalikes, audience overlap and recent posts.
Platforms and agencies building creator discovery into their own products.
Access and use
Needs an access token under a Modash API agreement; each call spends credits, published pricing starts at US$16,200 a year for Discovery and US$10,000 a year for Raw, and trial credits are offered after a call with the team. Modash also offers an MCP server for the same data. The API terms bar standalone resale of raw results and require a lawful basis for processing creator data.
A social data API that returns creators' public data and, with each creator's sign-in and consent, private data such as audience demographics and income from supported platforms, including YouTube.
Agencies and creator platforms that need data a creator has agreed to share.
Access and use
Sign up for a free sandbox, then talk to sales for production access and pricing. Consented data needs each creator to connect their account through Phyllo and approve the fields shared. Its developer documentation is published under the InsightIQ name at docs.insightiq.ai.
A YouTube channel database whose API searches over 35 million channels by niche, location, size and engagement and returns video lists and subscriber history, with sponsorship tracking in the web app.
Brands and agencies finding channels to sponsor and checking which creators competitors sponsor.
Access and use
The API uses a bearer key and prepaid credits, from 100 credits for US$24, charged per result, with free search previews and email lookups billed only when an email is found. Sponsorship tracking is part of the web app's Power and Enterprise plans, not the API. Business emails it returns are personal information, so outreach must follow spam and privacy law.
An Apify actor that collects public YouTube video and channel data, such as titles, views, likes, subscriber counts and upload dates, with optional subtitles.
Analysts who need bulk YouTube data without writing a scraper.
Access and use
Needs an Apify account and API token; it bills per video, about US$2.40 to US$4 per 1,000 depending on plan, and results are stored in your Apify datasets. The free plan's US$5 monthly credit covers a small run. It scrapes YouTube outside the official API, which YouTube's Terms of Service bar without permission, and Apify says not to collect personal data without a legitimate reason.
An open library from NVIDIA that builds an ontology from database metadata and serves it to agents through MCP, CLI and API, so they answer data questions with SQL that is checked against the database.
Data and platform teams who want agents to answer analytics questions over SQL sources such as PostgreSQL, Snowflake and Databricks using shared, certified definitions.
Access and use
Runs as a self-hosted stack: clone the repo, copy .env.example to .env, then docker compose up -d --build; a Helm chart covers Kubernetes. Requires Docker with Compose v2, uv and an NVIDIA API key (nvapi-...) from build.nvidia.com, or any OpenAI-compatible endpoint set per model role. By default the models that build the ontology and write each answer's SQL run on NVIDIA's hosted API catalog (integrate.api.nvidia.com); point the endpoints at self-hosted NIM to keep inference in house. Connect agents through the MCP query tools with uvx --from "git+https://github.com/NVIDIA/auto-ontology.git#subdirectory=mcp" auto-ontology-mcp; each user signs in, and every call runs with that user's application permissions. Generated SQL runs with the connected database account's rights, so use a dedicated read-only account limited to the tables and views you intend to expose. Compose publishes Postgres, pgAdmin and the web app on the host, so replace every default password first, and keep ports 3001 and 3002 on a private network because the ingestion endpoints have no authentication of their own. Setting KUMO_RFM_API_KEY makes the agent upload connected tables to the hosted KumoRFM service for predictive questions. Apache-2.0.
An open-weight OCR model family that parses Chinese and English PDFs and images into Markdown or JSON, with a second model that scores how reliable each recognised region is.
Teams building document agents or extraction pipelines who want OCR they can run on their own GPU, with a score that flags regions for human review.
Access and use
Install the SDK with python -m pip install --upgrade "minicpm-ocr[all]" (Python 3.10 or later) plus the PaddlePaddle GPU packages for layout detection, then serve the recognition model with vLLM, SGLang, llama.cpp, Ollama or FlagOS. The tutorial's vLLM route needs Linux x86_64, an NVIDIA GPU with driver 580 or later and at least 40 GB of disk, and it builds vLLM from source; the Docker image needs Linux x86-64, an NVIDIA GPU, a CUDA 12.9 compatible driver and the NVIDIA Container Toolkit. Deployed locally, document processing and inference stay on your machine, and the tutorial starts vLLM on 127.0.0.1; the first install and model downloads need internet. The free hosted API and online demo run at platform.modelbest.ai, so documents sent there leave your machine. The README suggests telling a coding agent to "Read and execute" a remote SKILL.md that sets up environments and downloads models: read it yourself before handing it over. Apache-2.0, covering code and model weights.