AI Admin API MCP Server
A read-only MCP server for querying AI provider administration APIs, providing normalized usage, cost, and dashboard data for OpenAI and Anthropic.
README
AI Admin API MCP Server
Read-only MCP server for querying AI provider administration APIs and returning normalized usage, cost, and dashboard data.
Status
Implemented providers:
- OpenAI Admin API usage, costs, projects, users, and project API keys.
- Anthropic Admin API organization metadata, workspaces, API keys, messages usage, and costs.
Planned providers:
- Google Cloud Billing export through BigQuery. This provider is not implemented yet. When it ships, release notes must state that direct BigQuery billing export queries can incur Google Cloud query costs.
All MCP tools are read-only. The server does not expose provider mutation tools.
Tech Stack
- Node.js 22+
- TypeScript
- pnpm
@modelcontextprotocol/sdk- Zod input schemas
- Vitest
Quick Start
Run the published package with npx:
OPENAI_ADMIN_KEY=sk-admin-... \
ANTHROPIC_ADMIN_KEY=sk-ant-admin-... \
npx @amaretto-software-labs/ai-admin-api-mcp --stdio
Pin a specific published version:
npx @amaretto-software-labs/ai-admin-api-mcp@0.0.1 --stdio
For local repo development:
pnpm install
pnpm check
pnpm test
Run over STDIO:
OPENAI_ADMIN_KEY=sk-admin-... \
ANTHROPIC_ADMIN_KEY=sk-ant-admin-... \
pnpm start:stdio
Run over Streamable HTTP:
OPENAI_ADMIN_KEY=sk-admin-... \
ANTHROPIC_ADMIN_KEY=sk-ant-admin-... \
MCP_HTTP_AUTH_TOKEN=local-proxy-token \
pnpm start:http
HTTP listens on http://127.0.0.1:8787/mcp. Pass --unsafe-local-http only for local development where another process already protects the endpoint.
Run the gateway-compatible local HTTPS endpoint with a local development certificate:
OPENAI_ADMIN_KEY=sk-admin-... \
ANTHROPIC_ADMIN_KEY=sk-ant-admin-... \
MCP_HTTP_AUTH_TOKEN=local-proxy-token \
MCP_HTTPS_CERT_PATH=/path/to/localhost.pem \
MCP_HTTPS_KEY_PATH=/path/to/localhost.key \
pnpm start:https
HTTPS listens on https://127.0.0.1:8787/mcp.
Use the published package for Streamable HTTP or HTTPS by passing the same flags to npx:
MCP_HTTP_AUTH_TOKEN=local-proxy-token \
OPENAI_ADMIN_KEY=sk-admin-... \
npx @amaretto-software-labs/ai-admin-api-mcp --http --port 8787
MCP_HTTP_AUTH_TOKEN=local-proxy-token \
OPENAI_ADMIN_KEY=sk-admin-... \
MCP_HTTPS_CERT_PATH=/path/to/localhost.pem \
MCP_HTTPS_KEY_PATH=/path/to/localhost.key \
npx @amaretto-software-labs/ai-admin-api-mcp --https --port 8787
Configuration
| Variable | Required | Description |
|---|---|---|
AI_ADMIN_ENABLED_PROVIDERS |
No | Comma-separated openai,anthropic. Defaults to providers with static credentials present. |
AI_ADMIN_REQUIRED_PROVIDERS |
No | Comma-separated providers that must be enabled and statically configured at startup. |
AI_ADMIN_CREDENTIAL_MODE |
No | Only static is implemented in this runtime build. pass_through and hybrid are documented gateway contracts and fail fast. |
OPENAI_ADMIN_KEY |
OpenAI static mode | OpenAI Admin API key. |
OPENAI_BASE_URL |
No | Override for tests or compatible OpenAI Admin API gateways. Defaults to https://api.openai.com/v1. |
ANTHROPIC_ADMIN_KEY |
Anthropic static mode | Anthropic Admin API key, sent as x-api-key. |
ANTHROPIC_OAUTH_TOKEN |
Anthropic static mode | Optional Anthropic OAuth bearer token with admin scope. |
ANTHROPIC_BASE_URL |
No | Override for tests or compatible Anthropic Admin API gateways. Defaults to https://api.anthropic.com/v1. |
ANTHROPIC_VERSION |
No | Anthropic API version. Defaults to 2023-06-01. |
ANTHROPIC_BETA |
No | Comma-separated Anthropic beta headers, for example fast-mode-2026-02-01. |
MCP_HTTP_AUTH_TOKEN |
HTTP mode | Bearer token required by the MCP HTTP endpoint unless unsafe local mode is used. |
MCP_HTTPS_CERT_PATH |
HTTPS mode | PEM certificate path for local HTTPS. |
MCP_HTTPS_KEY_PATH |
HTTPS mode | PEM private-key path for local HTTPS. |
MCP_CACHE_TTL_SECONDS |
No | Dashboard bundle cache TTL in seconds. Metadata list tools cache for 300 seconds. Defaults to 60. |
MCP_USER_AGENT |
No | Optional user agent context for future outbound request metadata. |
Credentials
Provider credentials must come from environment variables or an external host secret integration. They must never be sent as normal MCP tool arguments.
In v0.1 static mode, these credential refs are accepted:
credential:openai:staticcredential:anthropic:static
Omit credential_ref to use the static provider credential. Unknown refs are rejected.
Pass-through and broker modes are deployment contracts for an operator-supplied gateway. The gateway, credential store, tenant policy, audit layer, and envelope signing are outside this runtime build. See pass-through gateway docs.
Tools
Common tools:
ai_admin_list_providersai_admin_query_usageai_admin_query_costsai_admin_query_dashboard_bundle
OpenAI tools:
openai_admin_query_usageopenai_admin_query_costsopenai_admin_list_projectsopenai_admin_list_usersopenai_admin_list_project_api_keysopenai_admin_query_dashboard_bundle
Anthropic tools:
anthropic_admin_get_organizationanthropic_admin_list_workspacesanthropic_admin_list_api_keysanthropic_admin_query_messages_usageanthropic_admin_query_costsanthropic_admin_query_dashboard_bundle
Resources and Prompts
Resources:
ai-admin://providersai-admin://providers/{provider}/capabilitiesai-admin://schema/provider-capability-v1ai-admin://schema/usage-fact-v1ai-admin://schema/cost-fact-v1ai-admin://schema/dashboard-bundle-v1openai-admin://capabilitiesanthropic-admin://capabilities
Prompts:
build_usage_dashboardinvestigate_cost_spikeexport_finance_report
Provider Notes
- OpenAI usage endpoints have endpoint-specific filters, groupings, and metrics. Unsupported parameters are rejected before any provider call.
- Anthropic costs are reported in minor units and normalized to USD major units. Priority Tier costs are not included in the Anthropic cost endpoint.
- Anthropic
speedusage filters/groupings requireANTHROPIC_BETA=fast-mode-2026-02-01. - Google Cloud Billing support is planned only. BigQuery queries against billing export tables can be cost-bearing.
MCP Client Example
{
"mcpServers": {
"ai-admin-api": {
"command": "pnpm",
"args": ["--dir", "/absolute/path/to/ai-admin-api-mcp", "start:stdio"],
"env": {
"OPENAI_ADMIN_KEY": "sk-admin-...",
"ANTHROPIC_ADMIN_KEY": "sk-ant-admin-..."
}
}
}
}
Docker
docker build -t ai-admin-api-mcp .
docker run --rm -p 8787:8787 \
-e OPENAI_ADMIN_KEY \
-e ANTHROPIC_ADMIN_KEY \
-e MCP_HTTP_AUTH_TOKEN \
ai-admin-api-mcp
The Docker image starts Streamable HTTP mode by default.
Development
pnpm check
pnpm test
pnpm test:live:openai
pnpm test:live:anthropic
Live tests are skipped unless provider credentials are present.
Release
The Build and Publish GitHub Actions workflow runs pnpm check and pnpm test on pushes and pull requests targeting main.
Releases are manual through Actions -> Build and Publish -> Run workflow on main.
Inputs:
package-version: optional SemVer override. If omitted, the workflow uses the committed package version.npm-tag: npm dist-tag to use when publishing. Defaults tolatest.publish-to-npm: set totrueto publish@amaretto-software-labs/ai-admin-api-mcpto npm after the release is created.
The release job aligns the @amaretto-software-labs/ai-admin-api-mcp package version in the runner, rebuilds and tests the package, packs one npm tarball, pushes a v<package-version> git tag, creates a GitHub Release with the packed artifact, and optionally publishes @amaretto-software-labs/ai-admin-api-mcp to npm using trusted publishing. Configure npm trusted publishing for this repository and package before enabling publish-to-npm.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.