How is this different from connecting Replicate MCP directly?

Connected directly, an agent sees every tool the server exposes and holds the credential itself. Through TrustGate you choose which tools each agent gets, and every call is logged against the agent that made it.

What if the Replicate MCP API changes?

We track the Replicate MCP schema, so the tool catalogue stays current. Tools you have already allowed keep working, and anything new stays denied until you enable it.

Can I give different agents different Replicate tools?

Yes. Access is set per agent: one can get a handful of read tools, another the full set. You grant the whole MCP or single tools within it.

Where can I see what my agents did in Replicate?

Every call is logged against the agent that made it, with the tool it used. Activity can be sent on to your own monitoring.

Do I need to run the Replicate MCP server myself?

No. TrustGate connects to the Replicate MCP server and sits in front of it. Your agent only talks to your TrustGate URL, never to Replicate directly.

Connect Replicate MCP to your AI agents, securely.

Q: Is Replicate MCP secure?

The server itself exposes 37 tools, and some of them can write to your Replicate account. Behind TrustGate an agent only sees the tools you allow, requests and responses are inspected, and every call is logged. Anything outside the allowlist is denied before it reaches Replicate.

One secured endpoint for every Replicate call your AI agents make, scoped per user, inspected, and logged.

Get started for free

Get a demo

Try Replicate MCP now

What your agents can ask once Replicate MCP is connected.

Select the tools your AI agents can call

37 tools from the Replicate MCP schema.

Showing 37 tools

list_modelsreadGet a paginated list of public models.
search_modelsreadGet a list of public models matching a search query.
get_modelsreadGet a model, including its latest version and default example.
get_models_readmereadGet the README content for a model.
list_models_versionsreadList the versions of a model, sorted with the most recent version first.
get_models_versionsreadGet a model version, including the OpenAPI schema that describes its inputs and outputs.
list_collectionsreadGet a paginated list of collections of models.
get_collectionsreadGet a collection with a nested list of the models in that collection.
create_modelswriteCreate a model.
delete_modelshigh riskDelete a model. You can only delete private models you own that have no versions associated with them.
delete_models_versionshigh riskDelete a model version and all associated predictions, including all output files. Only versions of private models you own, not used by others, a training or a deployment, can be deleted.
create_predictionswriteCreate a prediction for the model version and inputs you provide.
create_models_predictionswriteCreate a prediction using an official model.
create_deployments_predictionswriteCreate a prediction for the deployment and inputs you provide.
get_predictionsreadGet the current state of a prediction.
list_predictionsreadGet a paginated list of all predictions created by the user or organization associated with the provided API token, including predictions created from the API and the website.
cancel_predictionswriteCancel a prediction that is currently running.
create_deploymentswriteCreate a new deployment with a model, version, hardware and minimum and maximum instances.
update_deploymentswriteUpdate properties of an existing deployment, including hardware, min/max instances, and the deployment's underlying model version.
list_deploymentsreadGet a list of deployments associated with the current account, including the latest release configuration for each deployment.
get_deploymentsreadGet information about a deployment by name including the current release.
delete_deploymentshigh riskDelete a deployment. You can only delete deployments that have been offline and unused for at least 15 minutes.
create_fileswriteCreate a file by uploading its content and optional metadata.
list_filesreadGet a paginated list of all files created by the user or organization associated with the provided API token.
get_filesreadGet the details of a file.
download_filesreadDownload a file by providing the file owner, access expiry, and a valid signature.
delete_fileshigh riskDelete a file. Once a file has been deleted, subsequent requests to the file resource return 404 Not found.
create_trainingswriteStart a new training of the model version you specify.
list_trainingsreadGet a paginated list of all trainings created by the user or organization associated with the provided API token.
get_trainingsreadGet the current state of a training.
cancel_trainingswriteCancel a training.
get_accountreadReturns information about the user or organization associated with the provided API token.
list_hardwarereadList the hardware available for running models and deployments.
get_default_webhooks_secretreadGet the signing secret for the default webhook endpoint. This is used to verify that webhook requests are coming from Replicate.
list_api_endpointsreadLists or searches the endpoints in the Replicate TypeScript API.
get_api_endpoint_schemareadGets the schema for an endpoint in the Replicate TypeScript API.
invoke_api_endpointhigh riskInvokes an endpoint in the Replicate TypeScript API.

What Replicate MCP exposes

Replicate MCP groups its 37 tools into 7 categories, each with a suggested starting policy.

All 37 tools as JSON, ready to paste.

Category	Covers	Access	Suggested policy
Models	Listing, searching and reading public models, their READMEs, examples, versions and collections, and creating or deleting the account's own models and model versions	Read / Write	Discovery here is read-only and safe to allow. Deleting a model version also deletes all of its predictions and output files, so delete_models and delete_models_versions belong behind approval
Predictions	Running a prediction on a model version, an official model or a deployment, reading a prediction's state, listing the account's predictions and cancelling a running one	Read / Write	Every create call runs a model on Replicate hardware billed to the account, so this is where spend comes from. Limit which models or deployments an agent may run, and note that list_predictions returns inputs and outputs from both the API and the website
Deployments	Creating, updating, listing, reading and deleting deployments, including their hardware, minimum and maximum instances and model version	Read / Write	update_deployments changes hardware and instance counts, which drives always-on cost, so keep create and update to infrastructure agents. Deletion only works on deployments idle for 15 minutes but is still irreversible
Files	Uploading files with metadata, listing and reading the account's files, downloading a file with a signed URL and deleting files	Read / Write	Files are account-wide, so reading them can expose inputs uploaded by other workflows. Treat delete_files as irreversible and gate it
Trainings	Starting a training of a model version, listing and reading trainings, and cancelling one	Read / Write	A training is a long-running, billed job that creates a new model at its destination, so create_trainings should be limited to agents explicitly meant to fine-tune
Account	Reading the account tied to the API token, listing available hardware and reading the default webhook signing secret	Read	Read-only, but get_default_webhooks_secret returns the secret used to verify webhooks from Replicate. Anyone holding it can forge webhook calls to your endpoint, so block it for agents that do not manage webhooks
API Endpoints	Listing and searching API endpoints, reading an endpoint schema and invoking any endpoint	Read / Write	invoke_api_endpoint can call any endpoint, including deletions; restrict it

Connect Replicate MCP to your AI agent

Four steps, under ten minutes.

01
Sign in to NeuralTrust
Open the NeuralTrust platform, sign in and go to Getting started, then Connect your tools.
02
Choose your agent
When asked which agent you use, pick Claude Code, Codex, Cursor, Windsurf, Copilot or your own code.
03
Connect Replicate
Find Replicate in the tool list and connect it.
04
Connect your agent
Follow the steps on screen to add the config to your agent, then send the test prompt.

Permissions down to the single tool

You do not allow Replicate MCP or block it. You allow the exact tools each agent can call, one by one.

Replicate MCP	37 TOOLS AVAILABLE
Code review agent	4 tools, read only
Issue triage bot	9 tools
Platform team assistant	all 37 tools

Tool by toolConnecting the server is not the unit of access. Each tool is granted on its own.
Default denyA tool you have not enabled is never exposed to the agent.
Approval on writesTools that change or delete things can require a human to confirm.

Is Replicate MCP secure?

One endpoint instead of raw credentials. Every call is checked against policy and logged.

AI agentsConnect through your MCP client configuration.	NeuralTrustApply controls based on your configuration.	Replicate MCP37 documented tools, one per Replicate HTTP API operation. The remote server at mcp.replicate.com connects over SSE and asks for a Replicate API token in a web authentication flow; the local replicate-mcp package reads it from REPLICATE_API_TOKEN.
Coding assistants	Connection & credentialsper user	Models11 tools
Internal copilots	Tool access policiesdeny by default	Predictions6 tools
Workflow automations	Request & response inspectioninline	Deployments5 tools
Compatible MCP clients	Activity monitoringalerts and logs	All other tools15 tools