thegraph-mcp-skill
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill connects to the official Model Context Protocol (MCP) endpoint for The Graph at
https://subgraphs.mcp.thegraph.com/sse. This communication is intended for fetching subgraph metadata and executing queries, and it targets a well-known, established service. - [COMMAND_EXECUTION]: The skill utilizes the
uxccommand-line utility to perform administrative tasks such as setting credentials viauxc authand creating local command aliases withuxc link. These operations are used to configure the local environment for interacting with the remote MCP server. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external subgraphs and GraphQL schemas, representing a potential indirect prompt injection surface. The skill mitigates this by providing specific guidance on query structure and output parsing.
- Ingestion points: The agent reads subgraph descriptions, entity schemas, and query result data through the
thegraph-mcp-clitool. - Boundary markers: The skill instructs the agent to rely exclusively on structured JSON envelopes (
ok,kind,data) and explicitly avoid raw text output parsing. - Capability inventory: The agent can execute CLI commands (
uxc,thegraph-mcp-cli) and perform network operations to the specified MCP endpoint. - Sanitization: Instructions mandate using filters, limits, and minimal field selections to reduce the volume and potential impact of ingested external content.
Audit Metadata