Repository navigation
Releases
Releases · nestjs/mcp
Release 0.0.4
Bug fixes
Full Changelog: 0.0.3...0.0.4
Release 0.0.3
- fix: validate the json of a tool's output against its schema (ed6ea33)
- test: type the resource-errors fixture with the call context (44f79c6)
- fix: normalize schema-declared tool outputs (#6) (1f504a0)
- fix: resource result error sanitization (#8) (d32f6c1)
- revert: drop the agent transport of the call context (7de9c53)
Release 0.0.2
- feat: give agent calls the tool call id and an idempotency key (e2b22a5)
- feat: export the primitive metadata key for packages that find tools (80fc271)
- feat: give calls from the app's own agent a context of their own (e73d85c)
- chore(deps): update dependency vitest to v5.1.0 (#2) (592704a)
- chore(deps): update dependency oxlint to v1.91.0 (#1) (1031a71)
- chore(deps): update nest monorepo to v12.3.3 (#4) (763de7f)
Release 0.0.1
The first release of @nestjs/mcp: Model Context Protocol servers built from Nest providers.
- Primitives: tools, resources, resource templates and prompts are provider methods with a decorator each (
@Tool(),@Resource(),@ResourceTemplate(),@Prompt()), with schemas from Zod, ArkType, Valibot or a JSON Schema. - Enhancers: guards, interceptors, pipes and exception filters run on every call, with the execution context type
'mcp', andMcpExecutionContextreads the call in them. - Streamable HTTP: the endpoint is a route of your application on Express, Fastify or the Node adapter. Each request is served on its own, for clients of the 2026-07-28 and the 2025 protocol revisions.
- stdio:
McpStdioServerserves the same module from a standalone application context. - Authorization: the HTTP endpoint can be an OAuth 2.1 resource server, with a token verifier of your own, protected resource metadata (RFC 9728) and scopes per tool, resource or prompt.
- Testing:
@nestjs/mcp/testingconnects an MCP client to the application in memory. - Auth packages: with
@nestjs/authentication0.1.3 and@nestjs/authorization0.1.1 or later, their global guard,@CurrentUser()and@Can()policies work on tool calls.