io.github.bakhtiersizhaev/openevidence-mcp

The materials show it requires YOUR_API_KEY, which is a sensitive credential; the description also mentions connecting via authenticated browser sessions, so runtime handling may involve session-based authentication data as well. There is no clear documentation on secret storage, least privilege, or anti-leakage controls, so credential misuse and session leakage should be considered.

Although no remote host is declared in the listing, the stated function is to connect OpenEvidence to AI agents, so it is expected to make network requests to OpenEvidence-related services and may send user queries or context outward. Because the remote endpoints are not explicitly listed in the materials, network egress is not fully auditable, but there is no clear red flag indicating unrelated or suspicious destinations.

The system checks indicate that this tool executes code or runs a local process, which is a standard MCP capability. The available materials do not show requests for system privileges beyond its stated purpose, nor do they reveal obvious command-injection or persistence mechanisms, so caution is appropriate rather than risk.

The mention of authenticated browser sessions implies that the tool may touch browser-session-related data and process user inputs sent to OpenEvidence. The materials do not specify what local files it can read or write, whether it accesses browser cookies/profile data, or whether it retains logs, so the data access boundary is insufficiently defined and should be verified before deployment.

Positive factors include being listed in an official registry, having open-source code, and being updated within the last year, all of which reduce supply-chain risk. Caution remains appropriate because community adoption is very low (0 stars), no license is declared, and the README is missing, leaving limited audit context and usage constraints.