generated: '2026-08-12'
method: probed
status: published
source: >-
  https://mcp.feedly.com/mcp — probed 2026-08-12 with an anonymous JSON-RPC tools/list; corroborated by
  https://docs.feedly.com/article/820-what-is-mcp-feedly and
  https://feedly.com/new-features/posts/feedly-mcp-server-automate-cti-workflows-with-claude-and-the-feedly-threat-graph
description: >-
  Feedly operates a hosted, remote Model Context Protocol server over its Real-Time Threat Graph,
  marketed as the "Threat Graph MCP Server" and bundled with the Advanced Threat Intelligence plan. It
  is a first-party production endpoint, not a community wrapper: the host answers a proper OAuth 2.1
  challenge and publishes RFC 8414 authorization-server metadata.

server:
  name: feedly-threat-graph
  transport: http
  url: https://mcp.feedly.com/mcp
  vendor: Feedly, Inc.
  plan_requirement: >-
    Threat Intelligence "Advanced" plan. The pricing page lists "Threat Graph MCP Server" as an
    Advanced-tier feature; it is not included in Standard.

x-evidence:
  fetched: '2026-08-12'
  probe: 'POST {"jsonrpc":"2.0","id":1,"method":"tools/list"}'
  url: https://mcp.feedly.com/mcp
  http_status: 401
  content_type: application/json
  body: '{"error":"invalid_token","error_description":"Missing or invalid access token"}'
  www_authenticate: 'Bearer realm="OAuth", error="invalid_token", error_description="Missing or invalid access token"'
  interpretation: >-
    A 401 with a well-formed OAuth challenge is positive evidence the server exists and is correctly
    protected. It is not a failed probe — it is the expected anonymous response from a production MCP
    server, and it is what distinguishes a real endpoint from a marketing claim.

authorization:
  model: OAuth 2.1 (authorization code + PKCE)
  metadata_document: well-known/feedly-mcp-oauth-authorization-server.json
  issuer: https://mcp.feedly.com
  authorization_endpoint: https://mcp.feedly.com/authorize
  token_endpoint: https://mcp.feedly.com/token
  registration_endpoint: https://mcp.feedly.com/register
  revocation_endpoint: https://mcp.feedly.com/token
  grant_types: [authorization_code, refresh_token]
  code_challenge_methods: [S256, plain]
  dynamic_client_registration: true
  protected_resource_metadata: >-
    Not published. /.well-known/oauth-protected-resource returns 404 on mcp.feedly.com, so an MCP
    client must fall back to the WWW-Authenticate challenge and the authorization-server metadata
    rather than discovering the AS from the resource per RFC 9728.

tools:
  count_documented: 16
  enumerated: false
  note: >-
    Feedly's own documentation states "the 16 tools exposed by the Threat Graph MCP server" but does
    not name them, and the live tools/list is OAuth-gated, so the real tool names and inputSchemas
    could not be read. NO TOOL LIST IS ASSERTED HERE. Deriving one from the REST operations would
    fabricate the provider's agent surface; the honest record is that the count is published, the
    schemas require authenticated introspection, and the capability areas below are the ones Feedly
    describes in prose.
  capability_areas_documented:
  - threat actors
  - malware families
  - vulnerabilities / CVEs
  - TTPs (MITRE ATT&CK techniques and procedures)
  - indicators of compromise
  - cyberattacks
  - trending / threat landscape
  source_for_areas: >-
    https://feedly.com/new-features/posts/feedly-mcp-server-automate-cti-workflows-with-claude-and-the-feedly-threat-graph
    and https://docs.feedly.com/article/823-feedly-mcp-server-demo

graph_scale_claimed:
  threat_actors: 800+
  malware_families: 10K+
  cves: 300K+
  iocs: 500K+
  techniques_and_tactics: 800+
  cyber_attacks: 10K+
  source: https://feedly.com/new-features/posts/feedly-mcp-server-automate-cti-workflows-with-claude-and-the-feedly-threat-graph
  note: Provider marketing figures, recorded as claims — not independently verified.

clients_documented:
- Claude (claude.ai, Claude Desktop, Claude Code)

related:
  rest_api: openapi/
  crosswalk: mcp/feedly-tool-crosswalk.yml
  agent_skills: skills/