CipherOwl for AI assistants

CipherOwl's remote Model Context Protocol (MCP) server lets an AI assistant
screen blockchain addresses, explain risk, and retrieve compliance metadata
using your existing CipherOwl organization and permissions.

Connection details

Environment	Streamable HTTP endpoint
Production	`https://mcp.cipherowl.ai/mcp`
Development testing	`https://mcp.dev.cipherowl.ai/mcp`

The server implements stateless MCP 2026-07-28 over HTTPS. Connect through
the ChatGPT plugin or Claude connector interface; do not paste a CipherOwl API
key or access token into a prompt. The host opens CipherOwl's OAuth login,
requests organization selection and consent, and stores the resulting
short-lived connection token.

The production endpoint remains unavailable until CipherOwl completes the
published connector review. Use the development endpoint only when CipherOwl
has supplied test access.

Tools and permissions

Every submitted tool is read-only and safe to retry. Screening through MCP
always disables alert tracking.

Tool	Purpose	OAuth scope
`detect`	Classify an address format locally	`detect:use`
`capabilities`	List the organization's supported capabilities	`capabilities:read`
`screen`	Screen one address	`screen:read`
`batch_screen`	Screen 1–500 addresses on one chain	`screen:read`
`reason_risk`	Explain risk categories and evidence	`risk:read`
`reason_exposures`	Explain counterparty exposures	`risk:read`
`metadata_labels`	Retrieve address labels	`metadata:read`
`metadata_entities`	Retrieve entity attribution	`metadata:read`

CipherOwl checks the token's issuer, audience, expiry, scopes, user, and
organization on every request. API-backed tools exchange it for a narrower
organization-bound token; the MCP token is never forwarded to the CipherOwl
API. Removing the connector or revoking access prevents future calls.

Good starter prompts

"Detect the chain for 17mhyeBX617ABZ1ffThhUTJkHUcMvCkfd5, then screen it."
"Screen 0x93df2c9c8786242b8fdca3f96b3914b8b4a9f704 and explain any risk."
"Show labels and entity attribution for this address before explaining its exposures."
"List my available capabilities before choosing a screening workflow."

Treat results as compliance decision support, not legal advice or an automatic
approval. Review material risk findings and ambiguous address/chain matches
before acting.

Data and support

Tool inputs and results are processed under CipherOwl's
Privacy Policy and
Terms. The server sends only the requested
address, chain, risk configuration, and operation to the applicable CipherOwl
API. It does not expose write tools, create cases, or enable alert tracking.

For connection errors, include the timestamp in the
CipherOwl support form. Never send an
OAuth token, client secret, or complete authorization header.