102 integrations, ready to install
102 maintained connectors across 20 categories, 1143 named actions between them. Your agents reach all of it through one MCP endpoint, behind the policy you set.
AI tools reached as rest apis
Showing 8 of 102 · Clear filters
Anthropic
8 actionsREST APIProvider REST API
Call Claude models, count tokens before you spend them, and read organization usage and cost.
ElevenLabs
19 actionsREST APIProvider REST API
Speak text in any voice, and generate images and video from a prompt.
OpenAI
10 actionsREST APIProvider REST API
Call models, embeddings, files, and vector stores.
OpenRouter
8 actionsREST APIProvider REST API
Compare models and prices, then call any of them with one key.
Google Gemini
9 actionsREST APIProvider REST API
Generate content with Gemini models, count tokens, and create embeddings.
Higgsfield AI
5 actionsREST APIProvider REST API
Generate images and video from a prompt, then poll until the file is ready.
Perplexity
11 actionsREST APIProvider REST API
Ask a question and get an answer with the sources it came from, synchronously or as a queued job.
Replicate
11 actionsREST APIProvider REST API
Find a model on Replicate, run it, and poll the prediction until the output is ready.
Four ways a connector reaches its provider
The catalog is not one adapter with 102 config files. A connector talks to whatever its provider actually ships, and where the call runs is what a call costs.
- Remote MCP server2 connectors · 1 credit per call
- The provider runs a streamable HTTP MCP server. The gateway calls it directly and forwards nothing else.
- Sandboxed MCP server3 connectors · 5 credits per call
- A stdio MCP server is started inside the project sandbox, spoken to over its standard streams, and torn down.
- Provider REST API92 connectors · 1 credit per call
- Each action is pinned to one named HTTP endpoint declared in the connector, signed with your stored credential at the moment of the call.
- Sandboxed CLI5 connectors · 5 credits per call
- The vendor's own command line tool runs inside the project sandbox, with secrets passed through the environment rather than the argument list.
Choose what each agent can call
Install a connector, decide who may call which of its actions, and read back the exact request and response for every call an agent made.