invokeable()

API & MCP

Programmatic access to the directory — on-demand lookups for humans, scripts, and agents. No API key, no SDK to install: plain HTTPS + JSON. Copy any sample below and it runs as-is. Machine-readable index: /api/v1 · openapi.json.

01REST API · one sample per endpoint, in your language

GET /api/v1/models

thin list of every model — ids, names, creators (details are per-entry, on demand)

language
shell
curl "https://invokeable.ai/api/v1/models"

GET /api/v1/models/{id}

everything about one model: every provider that serves it with that provider's exact model ID, pricing, benchmarks — and its source references (≥3 per model)

language
shell
curl "https://invokeable.ai/api/v1/models/claude-sonnet-4-5"

GET /api/v1/agents · /api/v1/agents/{id}

the same for agents: interfaces, install commands, invoke methods

language
shell
curl "https://invokeable.ai/api/v1/agents/claude-code"

GET /api/v1/providers · /api/v1/providers/{id}

auth model, base URL and docs link per provider

language
shell
curl "https://invokeable.ai/api/v1/providers/aws-bedrock"

GET /api/v1/snippet?model=&provider=&language=

working invocation code for model × provider × language, provider-specific model ID already substituted

language
shell
curl "https://invokeable.ai/api/v1/snippet?model=claude-opus-4-5&provider=aws-bedrock&language=python"

GET /api/v1/resolve?min_ctx=&capability=&modality=&max_in=&max_out=&provider=&creator=&sort=

constraints in, ranked concrete targets out — “vision, 1M context, under $2/M, on Vertex, cheapest first” becomes one call

language
shell
curl "https://invokeable.ai/api/v1/resolve?modality=image&min_ctx=1000000&max_in=2&provider=google-vertex&sort=price_in"

GET /api/v1/changes?since=YYYY-MM-DD&limit=

what's new — model releases since a date, newest first; poll it instead of scraping us

language
shell
curl "https://invokeable.ai/api/v1/changes?since=2026-08-01"

GET /api/v1/graph · /api/v1/graph/{id}

the relationship graph — typed nodes (model, agent, provider, creator, family, capability) and edges; /graph/{id} is one entity's neighborhood

language
shell
curl "https://invokeable.ai/api/v1/graph/claude-sonnet-4-5"

02MCP endpoint

Give your agent the directory as tools — it stops hallucinating model IDs and prices. Streamable HTTP, stateless, no auth: add https://invokeable.ai/mcp as a custom connector in Claude, or wire it into any MCP client.

search
find models/agents/providers by any identifier
get_model
full model entry incl. every provider's exact ID
get_agent
full agent entry incl. install + invoke code
get_snippet
invocation code for model × provider × language
whats_available
thin index of everything in the directory
language
shell
curl -X POST https://invokeable.ai/mcp \
  -H "content-type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_model","arguments":{"id":"claude-sonnet-4-5"}}}'

03Sources & provenance

Every model entry carries at least three source references in its sources field — the creator's official model docs, the creator's announcement surface, each serving provider's documentation, and the model's OpenRouter page where one exists. Benchmarks additionally cite their own source per row. If a number can't be traced, the entry says draft.

04Access policy

Everything here is free for on-demand use: look up a model, wire the API into your app, let your agent query it on every run. What we don't offer is a bulk dump — aggregating and keeping this fresh is the work, so wholesale copies aren't a download away. If you need broader access, talk to us.