Skip to main content

​
MCP tools reference

The @withfoundry/mcp-server exposes these tools to Cursor (and other MCP clients). All calls use your configured FOUNDRY_API_KEY and the API base URL https://api.withfoundry.ai.

​
Memory tools

​
foundry_memory_write

Write one or more entries into a namespace. Returns: { ids: string[] } — created entry ids. Example (as used by Cursor):
“Write to Foundry memory, namespace my-app: ‘We use API keys with prefix key_ in the Authorization header.’”
Semantic search over a namespace. Returns: Array of { id, content, metadata, score? }. Example:
“Search Foundry memory in my-app for ‘how do we authenticate?’”

​
foundry_memory_get_context

Get a single context string from semantic search for use in prompts. Returns: { text: string, entryIds: string[] }. Example:
“Get Foundry context for namespace my-app with query ‘user preferences and auth’, limit 3.”

​
Security tools

​
foundry_security_scan_repo

Start a security scan for a repository. Returns:
  • Without wait: { scanId, status }.
  • With wait: { scanId, status, findings }.
Example:
“Run a Foundry security scan on this repo” (Cursor may infer repo from workspace.)

​
foundry_security_get_scan_result

Get status and findings for a scan. Returns: { scanId, status, findings?, error? }. Example:
“Get Foundry scan result for scan_abc123.”

​
Envelope and errors

The MCP server calls the REST API. API responses use the envelope:
On failure, the tool returns the error from data (or a translated message) so Cursor can show or act on it.

​
Namespace tips in Cursor

  • Use a stable namespace per project or user (e.g. my-app, cursor-user-123).
  • Ask the AI to “search Foundry memory in namespace X for Y” so it uses the right namespace and query.