mcpField
B
backstage-mcp-server

backstage-mcp-server

Local
github.com/CoderrobDeveloper Tools3Updated 1d ago

About backstage-mcp-server

backstage-mcp-server exposes the Backstage Software Catalog to LLM clients as typed MCP tools. It is for teams integrating catalog discovery, validation, and maintenance into MCP workflows, with authentication and operational policies built in.

Key features

  1. 0132 Backstage Catalog tools
  2. 02Entity discovery and validation
  3. 03Catalog location and entity mutations
  4. 04Rotating bearer-token credentials
  5. 05Timeout, rate-limit, and cache policies
  6. 06Structured schemas and safe tool annotations

Use cases

  1. 01Query catalog entities and relationships from an MCP client
  2. 02Validate or refresh catalog entries
  3. 03Add or remove catalog locations and entities

MCP support

The stdio MCP server provides Backstage Catalog operations as tools for MCP clients, returning text and structured results. It validates inputs and applies authorization, caching, rate limits, and timeouts.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

No install method listed yet

We haven’t collected a package or remote endpoint for this server. Follow the instructions in the project README.

Repository

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

Backstage MCP Server

A type-safe Model Context Protocol server for the Backstage Software Catalog, built on the published @coderrob/mcp-kernel runtime.

It exposes 32 Catalog tools over stdio, uses Backstage's official Catalog client, supports rotating bearer credentials, and ships with black-box SDK and MCP Inspector verification.

Runtime architecture from an MCP client through the published MCP Kernel dependency and Backstage integration to the Catalog API.

What you get

CapabilityImplementation
Catalog discoveryEntity queries, references, ancestry, facets, and source locations
Catalog operationsAdd locations, refresh entities, remove entities or locations, and validate descriptors
Type safetyZod schemas infer handler inputs and emit MCP-compatible JSON Schemas
Runtime policiesTimeouts, caller scopes, per-principal rate limits, tagged caching, and invalidation
Safe operationsMCP annotations distinguish read-only, mutating, idempotent, and destructive tools
Production boundaryOfficial Backstage CatalogClient, authenticated fetch, redacted stderr logging
VerificationPer-file coverage, SDK stdio smoke tests, every-tool execution, and MCP Inspector

Quick start

1. Install and build

Requirements:

  • Node.js 24.15 or newer
  • Corepack
  • A reachable Backstage backend and an accepted external-access bearer token
corepack enable
corepack yarn install --immutable
corepack yarn build

The Yarn release comes from the packageManager field in package.json; use corepack yarn so local and CI behavior stay aligned.

2. Configure Backstage

Set the backend root or the complete Catalog plugin URL and choose one credential source:

export BACKSTAGE_BASE_URL=https://backstage.example.com
export BACKSTAGE_TOKEN=your-token

For externally rotated credentials, point to a token file instead:

export BACKSTAGE_BASE_URL=https://backstage.example.com
export BACKSTAGE_TOKEN_FILE=/run/secrets/backstage-token

BACKSTAGE_TOKEN_FILE takes precedence when both variables are present and is reread for every outgoing request. Optional LOG_LEVEL=debug enables diagnostic logging. Logs always go to stderr because stdout carries MCP protocol traffic.

3. Start the stdio server
corepack yarn start

The process waits for an MCP client on stdin/stdout. For a read-only check against a real deployment, use corepack yarn test:live; see live Backstage verification.

Connect an MCP client

After building, configure a stdio client with an absolute path:

{
  "mcpServers": {
    "backstage": {
      "command": "node",
      "args": ["/absolute/path/to/backstage-mcp-server/dist/cli.cjs"],
      "env": {
        "BACKSTAGE_BASE_URL": "https://backstage.example.com",
        "BACKSTAGE_TOKEN": "your-token"
      }
    }
  }
}

After global installation, use backstage-mcp-server as the command. Do not place credentials in committed client configuration; use the client's secret or environment mechanism where available.

Tool catalog

The checked-in tools-manifest.json is generated from the runtime definitions and is the machine-readable source of truth.

CategoryTools
Entity discoveryget_entities, get_entities_by_query, get_entities_by_refs, get_entity_by_ref, get_entity_ancestors, get_entity_facets
Location discoveryget_location_by_entity, get_location_by_ref
Validationvalidate_entity
Catalog mutationadd_location, refresh_entity
Destructive mutationremove_entity_by_uid, remove_location_by_id

get_entities.filter follows Backstage's key-value model: keys in one record are AND conditions, values for one key are OR conditions, and multiple records are OR groups.

{
  "filter": {
    "kind": ["Component", "API"],
    "metadata.namespace": "default"
  }
}

This selects Components or APIs in the default namespace. The Backstage integration guide documents every tool-to-client mapping, cursor behavior, authentication, permissions, and error conversion.

How it works

  1. An MCP client launches the packaged CLI and exchanges JSON-RPC over stdio.
  2. The published MCP Kernel validates and compiles immutable feature definitions.
  3. Zod validates input before middleware, authorization, caching, rate limiting, and timeout policies run.
  4. A Backstage tool calls the typed Catalog contract from its application context.
  5. The adapter delegates HTTP behavior to the official Backstage CatalogClient and injects the bearer token at the fetch boundary.
  6. Results return as MCP text plus structured content; expected failures use stable error codes and safe details.

The published @coderrob/mcp-kernel package owns protocol registration, lifecycle, policies, transports, testing support, and protocol-safe logging. This repository owns only the Backstage application and Catalog integration. See the current architecture overview for the dependency boundary.

MCP Kernel dependency

Generic server authoring, lifecycle, policies, transports, results, testing support, and logging are documented in the @coderrob/mcp-kernel repository. This package re-exports the kernel API for compatibility, but new generic consumers should install and import @coderrob/mcp-kernel directly.

To contribute a Catalog capability, follow Adding a Backstage MCP tool.

Verification

Verification layers from static quality checks through packaged black-box MCP Inspector validation.

Run the normal contributor gate:

corepack yarn lint
corepack yarn typecheck
corepack yarn architecture:check
corepack yarn knip
corepack yarn test
corepack yarn test:shell

For MCP, tool, adapter, or packaging changes, also run:

corepack yarn test:mcp
corepack yarn manifest:check

test:mcp runs contract tests, builds the distribution, launches dist/cli.cjs over stdio with the official SDK client, invokes every registered tool against a deterministic authenticated Catalog stub, and performs an independent MCP Inspector check. It does not bypass the server with direct handler or HTTP calls.

See Repository tooling for the purpose and side effects of every command and MCP end-to-end testing for the exact black-box boundary.

Documentation

Choose a path from the documentation hub:

License

GPL-3.0. See LICENSE.

Curated knowledge and skills

The Backstage knowledge base maintains classified articles, named source citations and schema.org metadata. Repository skills in .agents/skills support catalog lookups, entity maintenance and MCP development.

The 19 contextual lookup tools cover names, System and Domain contents, Groups, ownership, dependencies, APIs, annotations and orphaned entities. See lookup contracts and examples, including pagination, ambiguity and work bounds.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.