Documentation

MCP Server

The APIs.io catalog is also available to AI agents as a Model Context Protocol server over Streamable HTTP:

Code
https://apis.io/mcp

It's generated from the same OpenAPI contract as the REST API, so the two never drift.

Most tools read. The write tools all concern a listing you own or a request you are filing — claim_listing, correct_facts, dispute_finding, submit_artifact, generate_artifact, set_visibility, request_check, watch_listing / unwatch_listing. Every one of them returns 202 and is worked by a person: none publishes anything on your behalf or moves a score on its own. save_search / create_list / add_to_list write only to your own workspace.

You do not need a key to start. An anonymous client gets the whole free catalog surface — search, providers, APIs, tags, taxonomy, artifacts, the playground, and everything about one provider. Sign in only when you want the tools that synthesise across the catalog.

Install

Claude (web, desktop, mobile)

Settings → Connectors → Add custom connector, then:

Field	Value
Remote MCP server URL	`https://apis.io/mcp`
Name	`APIs.io`
OAuth Client ID / Secret	leave blank

Leave the OAuth fields empty — APIs.io supports dynamic client registration, so Connect registers your client for you and takes you through sign-in. The connector works before you authorize; you just get the free tier until you do.

Claude Code

Code
claude mcp add --transport http apis-io https://apis.io/mcp

Add --scope user to make it available in every project on your machine. Then /mcp to authorize.

Cursor

Add APIs.io to Cursor — or add it by hand to ~/.cursor/mcp.json:


Code
{
  "mcpServers": {
    "apis-io": {
      "url": "https://apis.io/mcp"
    }
  }
}

VS Code

Add APIs.io to VS Code — or MCP: Add Server → HTTP → https://apis.io/mcp.

Anything else

Any client that takes a remote MCP URL:


Code
{
  "mcpServers": {
    "apis-io": {
      "url": "https://apis.io/mcp"
    }
  }
}

The server speaks JSON-RPC over Streamable HTTP: initialize, tools/list / tools/call, prompts/list / prompts/get, and resources/list / resources/read. It negotiates protocol revision 2025-06-18. If your client shows prompts, start there — they chain the right tools for a job (see Prompts below).

Verify any install with one unauthenticated call:

Code
curl -s https://apis.io/mcp \
  -H 'content-type: application/json' \
  -H 'accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

Authentication

Anonymous callers get the free tier. To raise it, either sign in over OAuth or send a key:

OAuth 2.1 — authorization-code + PKCE (S256), with dynamic client registration and refresh tokens. Discovery is at /.well-known/oauth-protected-resource and /.well-known/oauth-authorization-server. Scopes: apis:read, apis:pro, apis:business, offline_access. Most clients need nothing but the URL. apis:pro is granted only to a signed-in person on Understanding or Influence and apis:business only on Influence; a self-registered client tops out at apis:read whatever it asks for (see onboarding).
API key — send x-api-key on the MCP endpoint, the same key the REST API uses.

Tools above your tier return upgrade_required rather than failing — see Plans.

The catalog itself is free: search, providers, apis, tags, taxonomy, artifacts, the playground, and everything about one provider. Anything that synthesises across the catalog — ratings, cohorts, capabilities, operations, insights, stack design, resolve/enrich — needs Understanding. Agent-readiness across the catalog, vendor comparison, gap analysis and the whole own-your-listing surface need Influence. report_correction and submit_feedback are free and unmetered: if the catalog is wrong about you, say so without paying.

Tools

apis_io_search is the power tool — find APIs, providers, or tags by free text, tags, providers, artifact types, industry, region, and rating band. The full set (tools needing Understanding or Influence marked 🔒):

Discovery

find_apis / get_api, find_providers / get_provider, find_tags / get_tag
find_artifacts — artifacts of a given type (OpenAPI, AsyncAPI, Postman, …) across the catalog
find_similar_providers / find_similar_apis, get_provider_artifacts / get_api_artifacts
get_openapi, get_provider_onboarding
resolve — turn any identifier (domain, URL, github.com/<org>) into the provider it belongs to (free)
enrich_provider — one call, pick field groups (profile, onboarding, artifacts free; rating, insights 🔒) instead of chaining several get_provider* calls

Venture Capital (VC firms as a first-class entity + their portfolio graph — free)

find_vcs — search VC firms by name, thesis tags, or category; sort by portfolio_on_network / portfolio_total / portfolio_rating
get_vc — one firm: identity, fund facts, portfolio summary (band distribution + average rating); view=full inlines the whole portfolio
vc_portfolio — a firm's portfolio companies, each network-matched + rated (filter is_provider=true for only API providers, or by band / min_score)
find_investors — the reverse edge: which VC firms back a given provider/company

Provider surface shortcuts — one call each, instead of walking a provider's artifacts

get_provider_operations — every operation across all of a provider's OpenAPIs: method, path, operationId, summary, which API it belongs to, deprecation. Filter by method, api, path, or free text. For a provider like Stripe this replaces fetching and parsing 159 specs — free
get_provider_tools — every MCP tool they ship, with the operation each wraps and its auth — free
get_provider_schema — every JSON Schema they publish, the data shapes their API works in — free

Two things these report that a raw artifact walk does not. Operations carry their agentic action class and consequence where the provider publishes agentic-access — the difference between a read and something that moves money — so ?consequence=read finds the safe surface before you point an agent at it. An operation with no agentic block was not profiled, which is not the same as safe, and meta.coverage says how many have it.

Tools carry provenance: first-party means the provider publishes that server, derived means we generated a candidate tool list from their OpenAPI because no hosted server was found. Use ?provenance=first-party for a true picture. A server whose tools could not be enumerated — auth-gated, say — is reported in coverage.not_enumerable with the reason rather than counted as zero.

Cohorts (any scored population of providers as data — a tag, industry, region, area, or VC portfolio)

find_cohorts — the index: 13,764 cohorts, filterable by kind, population, or promotion tier (report 100+, brief 25+, data) — free
get_cohort — one cohort's identity and full member roster, each with its Kin Score and agent readiness — free
cohort_stats 🔒 — the distribution for a whole market: mean/median/min/max/stdev, band split, per-facet averages, adoption rates. The market-stats sheet from a Trend Report, computed live
cohort_rankings 🔒 — the leaderboard on two axes: composite rank and agent rank. The two orders differ, and the gap between a provider's two ranks is usually the story
cohort_scores 🔒 — facet-level scores carrying the cohort average and the delta against it. A 60 in governance means nothing until you know the market sits at 45
cohort_capabilities 🔒 — what every member publishes, and which agent dimensions each satisfies
compare_cohorts 🔒 — two markets side by side on one rubric. A question a written report cannot answer, because a report covers one market

Every response carries a coverage block saying how much of the roster is actually scored — a mean over half a roster is a different claim from a mean over all of it. Cohorts below five scored members answer 422 rather than emit a statistic nobody should quote.

cohort_stats reports MCP adoption twice, on purpose: mcp_pct counts any mcp/ artifact, but a majority of those are ours — derived from the provider's own OpenAPI when no hosted server was found. mcp_first_party_pct counts only servers the provider publishes. The gap is large and consistent (payments: 32% against 9%). Use the first-party figure for any claim about what a market ships.

Provider score depth

get_provider_capabilities — artifact counts by type plus the agent dimensions satisfied — free
get_provider_evidence — how each part of a score was established: first-party (they published it), verified (we fetched and confirmed it), or derived (we inferred it). Ungated, deliberately — the basis for a claim should never sit behind the claim. Call it before quoting or disputing a score
get_provider_facets 🔒 — the facets the composite is built from, plus any regulatory regime that applies
get_agent_readiness 🔒 — the agent-readiness dimensions: spec presence, MCP server, auth clarity, idempotency, error semantics, rate-limit signal, well-known catalog, consent identity. A standalone score, not a slice of the composite

Saved Workspace (an Understanding dimension — persistent per-user state, requires a key) 🔒

save_search / list_saved_searches / run_saved_search / delete_saved_search
saved_search_net_new — what newly matches your saved criteria since you last checked (the agent-native demand report)
create_list / list_lists / get_list / add_to_list / delete_list — build and keep named shortlists/stacks
These are the same capability as the REST API's /me/* endpoints — the API and MCP share one implementation and one paywall, so they never drift.

Taxonomy — find_industries / get_industry, find_regions / get_region, find_areas / get_area are all free (browse + get); only the ranked get_industry_leaders / get_region_leaders / get_area_leaders are 🔒

Ratings — get_rating_rubric (free); find_ratings / get_provider_rating / get_rating_history / find_rating_movers 🔒

Synthesis — recommend_stack is freemium (free top pick per capability, Understanding for the full stack + alternatives + APIs.json); compare_providers, gap_analysis, industry_gap_analysis, whats_changed, export_stack 🔒

Per-type artifact discovery — one find_<type> per artifact type, all free except the two marked. Named in full because an agent scanning for a tool name will not expand an ellipsis:

find_apis_json · find_arazzo · find_asyncapis · find_channels · find_collections · find_examples · find_finops · find_graphql · find_json_ld · find_json_schemas · find_json_structures · find_mcp · find_openapis · find_plans · find_postman · find_rate_limits · find_rules · find_scopes 🔒 · find_security 🔒 · find_skills

Data products — export_dataset (the whole ratings/providers dataset in one pull, Influence license, with co_brand) and story_leads (owner-only editorial rollup)

Insights (demand-side: what the Fortune 1000 + API providers build, buy, and hire for)

Explore: find_company_insights, insights_dimensions, insights_adoption
Understanding 🔒: get_company_insight (full 40-dimension profile), company_gaps, match_providers (the supply↔demand join)

🔒 tools require an Understanding or Influence key (Plans). Call tools/list for the authoritative, always-current set:

Code
curl -s https://apis.io/mcp -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Prompts

Prompts are guided flows — each one chains the right tools for a job, so you don't have to know the tool names. Ungated prompts run for anyone; 🔒 prompts need an Understanding or Influence key and are listed for everyone (they upsell on use). Call prompts/list for the always-current set, prompts/get to run one.

Discover & adopt — find_api, provider_overview, explain_artifact, integrate_provider, api_readiness_check, agent_readiness_scan, best_in_class, explore_industry, explore_area, rating_methodology, insights_landscape, demand_vs_supply

Improve a provider you own — improve_my_score; claim_and_improve 🔒, benchmark_against_peers 🔒

Sell & go to market — sell_upmarket 🔒

Data & the "state of" story (media) — state_of_report 🔒 (Influence), movers_briefing 🔒, category_leaderboard 🔒

Decide — design_api_stack 🔒, vendor_shortlist 🔒, audit_api_estate 🔒, find_alternatives 🔒, build_agent_toolset 🔒, track_provider_changes 🔒, company_readiness 🔒

Sensemaking (owner) — story_leads, network_pulse 🔒

Code
# list prompts, then run one
curl -s https://apis.io/mcp -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"prompts/get","params":{"name":"improve_my_score","arguments":{"slug":"twilio"}}}'

Resources

Attachable, readable context an agent can pin into a session. Ungated unless marked 🔒.

apis://catalog, apis://ratings/rubric, apis://llms.txt — the network root, the scoring rubric, and the machine-readable index
apis://changes 🔒 — the recent catalog + rating movement feed
Templated: apis://provider/{slug}, apis://provider/{slug}/openapi, apis://tag/{slug} (free); apis://provider/{slug}/dossier, apis://industry|region|area/{slug}, apis://stack/{capabilities} 🔒

Call resources/list and resources/templates/list for the authoritative set.

See the API Reference for the underlying operations each tool maps to.

Last modified on September 16, 2026

Authentication Rate Limits