generated: '2026-09-19'
method: searched
status: published
source: https://machinelibrary.ai/mcp
sources:
- https://machinelibrary.ai/.well-known/mcp/server-card.json (saved verbatim: mcp/machinelibrary-ai-server-card.json)
- https://machinelibrary.ai/install.md (saved verbatim: skills/machinelibrary-ai-install.md)
- https://machinelibrary.ai/llms.txt (tool list)
- https://github.com/SpaceFrontiers/mcp (MIT source: tools.py, prompts.py, resources.py, server.json, registry.json, smithery.yaml, AGENTS.md)
- https://registry.modelcontextprotocol.io/v0/servers?search=spacefrontiers (io.github.SpaceFrontiers/mcp, versions 0.2.1 -> 0.3.1)
- https://api.machinelibrary.ai/auth.md
deployment:
  mode: both
  endpoint: https://mcp.machinelibrary.ai
  legacy_endpoint: https://mcp.spacefrontiers.org
  install: "git clone https://github.com/SpaceFrontiers/mcp.git && cd mcp && uv sync && SEARCH_API_ENDPOINT=https://api.machinelibrary.ai SPACE_FRONTIERS_API_KEY=<key> uv run fastmcp run mcp_server.py"
  package: https://github.com/SpaceFrontiers/mcp
  package_note: pyproject name spacefrontiers-mcp 0.3.1, but it is NOT published to PyPI (pypi.org/pypi/spacefrontiers-mcp 404 on 2026-09-19); the stdio path is a source checkout run with uv, exactly as the README's "Self-hosted (stdio)" section states.
  auth: oauth
  auth_detail: OAuth 2.1 authorization code + PKCE S256 with RFC 7591 dynamic client registration against https://api.spacefrontiers.org (scope search), OR a long-lived Bearer API key from https://machinelibrary.ai/keys; the WorkOS-style service_auth claim flow is offered for clients that cannot receive a browser callback (auth.md).
  verified: probed
  verification: >-
    POST https://mcp.machinelibrary.ai/ with tools/list and with initialize (Accept: application/json, text/event-stream)
    both returned HTTP 401 "Authentication required" with WWW-Authenticate: Bearer
    resource_metadata="https://mcp.machinelibrary.ai/.well-known/oauth-protected-resource" and x-ratelimit-limit: 60.
    GET / returns 405. The legacy mcp.spacefrontiers.org answers identically with its own resource_metadata URL.
    tools/list is therefore auth-gated; the tool inventory below is taken from the provider's published sources,
    and the input schemas from the provider's MIT-licensed server source (tools.py), not from a live introspection.
server:
  name: org.spacefrontiers/search
  registry_name: io.github.SpaceFrontiers/mcp
  title: Machine Library by Space Frontiers
  server_info_name: Machine Library MCP Server
  version: 0.3.1
  transport: streamable-http
  url: https://mcp.machinelibrary.ai
  stateless: true
  supported_protocol_versions: ['2025-03-26', '2025-06-18', '2025-11-25']
  capabilities:
    tools: true
    prompts: true
    resources: false
  capabilities_note: server-card says resources false, but the source registers one resource template (spacefrontiers://document/{uri_b64}); recorded as declared by the provider, discrepancy noted.
  listings:
  - Official MCP Registry (registry.modelcontextprotocol.io) as io.github.SpaceFrontiers/mcp — 0.2.1 (2026-05-07), 0.3.0 (2026-07-29), 0.3.1 (2026-09-12); repo server.json already says 0.3.2
  - Smithery (smithery.ai/servers/@SpaceFrontiers/mcp)
  - Glama (glama.ai/mcp/servers/SpaceFrontiers/mcp)
  - one-click install links for Cursor, VS Code, VS Code Insiders on https://machinelibrary.ai/mcp
  license: MIT
auth:
  methods: [oauth2, apikey]
  oauth2:
    authorization_server: https://api.spacefrontiers.org
    protected_resource_metadata: https://mcp.machinelibrary.ai/.well-known/oauth-protected-resource
    authorization_server_metadata: https://api.spacefrontiers.org/.well-known/oauth-authorization-server
    scopes_supported: [search]
    grant_types: [authorization_code, refresh_token, 'urn:ietf:params:oauth:grant-type:jwt-bearer', 'urn:workos:agent-auth:grant-type:claim']
    pkce: S256
    dynamic_client_registration: https://api.spacefrontiers.org/v2/oauth/register
    token_endpoint_auth_methods: [none]
    discovery: well-known/machinelibrary-ai-mcp-oauth-protected-resource.json
  apikey:
    header: Authorization
    scheme: Bearer
    key_prefix: sf_live_
    keys_page: https://machinelibrary.ai/keys
tool_count: 5
tools:
- name: spacefrontiers_search_documents
  category: search
  read_only: true
  idempotent: true
  annotations: {readOnlyHint: true, idempotentHint: true, openWorldHint: true, destructiveHint: false}
  description: Search papers, books, patents, standards, Wikipedia, and YouTube transcripts (the `documents` index). Defaults to 10 compact, reranked hits.
  inputSchema:
    source: tools.py (provider source, commit on main 2026-09-12); schema shape as FastMCP generates it from these annotated parameters
    type: object
    properties:
      query: {type: string, maxLength: 16384, default: '', description: Free-text search query; may be empty when at least one filter is supplied}
      limit: {type: integer, minimum: 1, maximum: 30, default: 10}
      offset: {type: integer, minimum: 0, exclusiveMaximum: 500, default: 0, description: offset + limit must not exceed 500}
      filter_issns: {type: [array, 'null'], items: {type: string}, description: Journal ISSN, hyphenated or plain}
      filter_types: {type: [array, 'null'], items: {type: string}, description: CrossRef-style document type (journal-article, book, book-chapter, proceedings-article, posted-content, patent)}
      filter_issued_after: {type: [integer, 'null'], description: Unix seconds UTC}
      filter_issued_before: {type: [integer, 'null'], description: Unix seconds UTC}
  outputSchema: SearchResults {query, index, hits[DocumentResult], count, total, has_more, next_offset}
  source_operation: openapi/machinelibrary-ai-openapi.yml#searchDocuments (index_names [documents])
- name: spacefrontiers_search_social
  category: search
  read_only: true
  idempotent: true
  annotations: {readOnlyHint: true, idempotentHint: true, openWorldHint: true, destructiveHint: false}
  description: Search Reddit, Telegram channels, and Discord (the `social` index). Defaults to 10 compact, reranked hits.
  inputSchema:
    source: tools.py
    type: object
    properties:
      query: {type: string, maxLength: 16384, default: ''}
      limit: {type: integer, minimum: 1, maximum: 30, default: 10}
      offset: {type: integer, minimum: 0, exclusiveMaximum: 500, default: 0}
      filter_uri_prefixes: {type: [array, 'null'], items: {type: string}, description: e.g. https://reddit.com/r/MachineLearning/ or @channel_username}
      filter_issued_after: {type: [integer, 'null']}
      filter_issued_before: {type: [integer, 'null']}
  outputSchema: SearchResults
  source_operation: openapi/machinelibrary-ai-openapi.yml#searchDocuments (index_names [social])
- name: spacefrontiers_fetch_document
  category: documents
  read_only: true
  idempotent: true
  annotations: {readOnlyHint: true, idempotentHint: true, openWorldHint: true, destructiveHint: false}
  description: Retrieve bounded full text (40K chars default, up to 100K) plus up to 50 references for one canonical URI (doi:, https://doi.org/, arxiv:, pmid:, isbn:, Reddit/Telegram/YouTube URLs). Citation backlinks are opt-in because they add a separately billed search.
  inputSchema:
    source: tools.py
    type: object
    required: [uri]
    properties:
      uri: {type: string, maxLength: 16384, description: Copy verbatim from a source_uri; do not guess}
      max_chars: {type: integer, minimum: 1000, maximum: 100000, default: 40000}
      referenced_by_limit: {type: integer, minimum: 0, maximum: 20, default: 0}
  outputSchema: FullDocument {id, title, source_uri, uris, abstract, content, content_truncated, full_content_length, authors, issued_date, document_type, publisher, languages, tags, references[], referenced_by[]}
  source_operation: openapi/machinelibrary-ai-openapi.yml#getDocumentByUri (+ searchDocuments when referenced_by_limit > 0)
- name: spacefrontiers_search_in_document
  category: documents
  read_only: true
  idempotent: true
  annotations: {readOnlyHint: true, idempotentHint: true, openWorldHint: true, destructiveHint: false}
  description: Return up to five scored passages (2K chars each) matching a query inside one document, without the full body.
  inputSchema:
    source: tools.py
    type: object
    required: [uri, query]
    properties:
      uri: {type: string, maxLength: 16384}
      query: {type: string, minLength: 1, maxLength: 16384}
  outputSchema: DocumentPassages {source_uri, title, passages[{text, score, field, chunk_id, truncated}], passage_count, message}
  source_operation: openapi/machinelibrary-ai-openapi.yml#getDocumentByUri (text_filter)
- name: spacefrontiers_top_up_balance
  category: payments
  read_only: false
  idempotent: false
  description: Fund the account balance through a Stripe MPP payment challenge (fixed USD packages). Hosted-only capability named in install.md and the README; not in the open-source tools.py.
  inputSchema:
    source: not published (hosted-only tool; live tools/list is auth-gated)
    note: "The REST equivalent takes {package: usd-5|usd-10|usd-25|usd-50|usd-100|usd-250|usd-500|usd-1000}."
  source_operation: openapi/machinelibrary-ai-openapi.yml#createMppBalanceTopUp
prompts:
- name: deep_research_agent
  description: A meticulous deep-research agent that finds high-quality, citable evidence and synthesizes accurate answers grounded in sources.
  tags: [research, academic, citations]
  source: prompts.py
resources:
- uri_template: spacefrontiers://document/{uri_b64}
  name: Space Frontiers document
  description: Canonical document URI, base64url-encoded; mirrors spacefrontiers_fetch_document for clients that browse resources.
  source: resources.py
runtime_notes:
  rate_limit_header_observed: x-ratelimit-limit 60 / x-ratelimit-remaining / x-ratelimit-reset on the MCP host (unauthenticated 401)
  billing: search $0.01 + $0.001 per returned result (10-hit default = $0.02); fetch $0.05; in-document search $0.015; referenced_by_limit > 0 adds a billed search (README pricing section, corroborated by /v1/pricing)
  errors: insufficient funds and auth failures surface as MCP isError results (tools.py _handle_billing_errors)
  origin_and_protocol_checks: auth.py enforces an Origin allowlist and an MCP-Protocol-Version allowlist