This server also ships inside Apollo GraphQL Skills — a plugin that installs it already configured.
Three tools need no account; nine can change your graph.
Apollo runs this one, so there is nothing to install: you add one URL and an agent gains the GraphOS control surface that rover gives a terminal. It answers the questions you would otherwise open Studio for — which launch broke composition, which client is driving the p99 latency, whether a schema change touches a live operation.
What it is
An MCP endpoint Apollo runs. Two unrelated things sit behind that one URL: Apollo's own documentation plus the Apollo Connectors specification, and the graph-management side of GraphOS — the schema registry, launches, schema checks and traffic data for the graphs your account can reach. It is 32 tools in total, and the client reads the catalogue as soon as the URL is added.
What you get
- Search across Apollo's official documentation — GraphQL, GraphOS, Apollo Router, Apollo Client, schema design, deployment and Connectors — and the full text of any page that search returns
- The Apollo Connectors specification, for writing or changing schemas that use
@connector@source - The graphs and variants a credential actually covers, resolved from the credential itself, so a question about "my graph" does not need an ID typed into it
- Published schemas at three widths: one subgraph with its routing URL and revision, the whole API schema, or the composed supergraph schema with its composition errors
- A variant's federation version, GraphQL endpoint URL and subgraph inventory, its README, a contract variant's include and exclude tag lists, and whether it has a persisted query list with its revision and operation count
- Launch history for a variant, and any single launch in full: status, which subgraphs changed, composition errors, and a diff against the previous launch with the operations it affects
- Schema checks from both ends — the list of past runs with their status, commit and subgraph, and one run's outcome: composition errors, lint diagnostics and the client operations each schema change touches
- Lint diagnostics with coordinate, severity, rule and source location, either from the graph's recent checks or for a schema document you hand in, which publishes nothing
- Client operations validated against a variant's published schema, each problem returned with its type, code and the operation it came from
- Traffic as CSV, ranked by request count, p99 latency or error count, broken down by operation, by subgraph or connector, and by client name and version
- Writes, nine of them: publish a subgraph, a monograph schema, a contract variant, a persisted query list or a variant README; start a schema check; delete a subgraph or a whole graph
- A dry run on subgraph deletion, which reports the composition result the removal would cause and deletes nothing
Requirements
An MCP client that speaks Streamable HTTP; Apollo's install page covers Claude, Claude Code, Cursor, VS Code, Codex CLI, Gemini CLI and the GitHub Copilot coding agent. The three documentation tools need nothing else — no account and no key. The other 29 need a GraphOS identity, and there are two ways to give it one: sign in to GraphOS through the client's own OAuth flow, which Apollo recommends because it leaves no key to store or rotate, or send a GraphOS API key in an x-api-key header when the client cannot open a browser, such as an agent running in CI. A personal API key reaches every graph your account can access; a graph API key is scoped to a single graph. Either credential is forwarded to the Apollo Platform API, which decides what the agent is allowed to do.
Setup effort
Paste a URL, then authorize — add the endpoint to your client, then approve the OAuth consent screen