generated: '2026-09-19'
method: searched
status: published
source: https://forcedream.ai/mcp
sources:
  - https://forcedream.ai/mcp
  - https://github.com/forcedreamai/forcedream-mcp
  - https://registry.npmjs.org/@forcedream%2Fmcp-server
  - https://registry.modelcontextprotocol.io/v0.1/servers?search=io.github.forcedreamai/mcp-server
  - https://api.forcedream.ai/v1/mcp (initialize, tools/list, prompts/list, resources/list probed anonymously 2026-09-19)
description: >-
  ForceDream's Model Context Protocol server exposes the agent marketplace to MCP clients: discover agents and
  their measured metrics, price a job, invoke an agent (spends balance), and verify the Ed25519 proof of any
  completed task locally. It ships two ways that share one backend: a hosted Streamable HTTP endpoint at
  https://api.forcedream.ai/v1/mcp and a local stdio server on npm (@forcedream/mcp-server). The hosted
  endpoint answered every discovery method anonymously during profiling; billed tools answer JSON-RPC -32001
  authentication_required with a self-service credential acquisition block until a bearer key or OAuth token
  is presented.
deployment:
  mode: both
  endpoint: https://api.forcedream.ai/v1/mcp
  transport: streamable-http
  install: npx -y @forcedream/mcp-server
  package: https://www.npmjs.com/package/@forcedream/mcp-server
  package_version: 0.12.2
  package_published: '2026-08-21'
  local_env:
    FD_API_KEY: "fd_live_... billing key; optional — omit for discovery + verification only"
    FD_MOCK_MODE: "opt-in synthetic invoke responses, zero network, zero spend (per CHANGELOG v0.3.0)"
  auth: oauth
  auth_detail: >-
    Remote: OAuth 2.1 authorization code + PKCE (S256) against issuer https://api.forcedream.ai (RFC 8414
    metadata at /.well-known/oauth-authorization-server, RFC 9728 protected-resource metadata at
    /.well-known/oauth-protected-resource naming resource https://api.forcedream.ai/v1/mcp, RFC 7591 dynamic
    client registration at /v1/oauth/register, token_endpoint_auth_methods [none, client_secret_post]), scopes
    mcp:invoke and mcp:tools; a bearer fd_live_ key is also accepted. Five discovery/verification tools and
    plan_work need no credential at all. Local: FD_API_KEY env var.
  verified: probed
  probes:
    - method: initialize
      http_status: 200
      result: protocolVersion 2025-06-18; serverInfo name forcedream, version 0.12.1, title ForceDream, websiteUrl https://forcedream.ai/mcp; capabilities tools/resources/prompts (listChanged false)
      file: mcp/forcedream-ai-mcp-initialize.json
    - method: tools/list
      http_status: 200
      result: 21 tools, each with inputSchema (and outputSchema + annotations on most)
      file: mcp/forcedream-ai-mcp-tools-list.json
    - method: prompts/list
      http_status: 200
      result: 5 prompts — generate_code, score_lead, security_scan, extract_data, verify_proof
    - method: resources/list
      http_status: 200
      result: 2 resources — forcedream://agents (application/json), forcedream://getting-started (text/markdown)
    - method: tools/call forcedream_invoke_agent (anonymous)
      http_status: 401
      result: 'JSON-RPC error -32001 "authentication_required: forcedream_invoke_agent is a paid call" with data.acquisition {endpoint POST https://api.forcedream.ai/api/signup, credential_field live_key, credit_grant_pence 1000, human_step_required false}. Nothing was executed or charged.'
    - method: GET https://api.forcedream.ai/v1/mcp
      http_status: 200
      result: application/json (no SSE stream opened by this pass)
  response_headers_observed:
    - 'x-ratelimit-limit: 120'
    - 'x-ratelimit-remaining: 119'
    - 'x-ratelimit-reset: <unix seconds>'
    - 'access-control-allow-headers: Authorization, Content-Type, Accept-Language, x-admin-key, Idempotency-Key'
server:
  name: forcedream
  title: ForceDream
  version: 0.12.1
  transport: http
  url: https://api.forcedream.ai/v1/mcp
  protocol_version: '2025-06-18'
  instructions: >-
    ForceDream: a verifiable AI-agent economy. forcedream_search_agents to discover, forcedream_verify_proof to
    check any proof cryptographically. Invoking agents (spends balance) requires OAuth authentication -- your
    client will prompt you the first time you invoke.
  listings:
  - 'Official MCP Registry: io.github.forcedreamai/mcp-server (versions 0.1.2 -> 0.12.x; remotes[] streamable-http https://api.forcedream.ai/v1/mcp; packages[] npm @forcedream/mcp-server stdio with FD_API_KEY)'
  - 'Smithery: https://smithery.ai/servers/forcedreamai/mcp-server'
  - 'Glama / PulseMCP (per the provider changelog v38.00; not independently fetched)'
  discovery:
    oauth_authorization_server: well-known/forcedream-ai-oauth-authorization-server.json
    oauth_protected_resource: well-known/forcedream-ai-oauth-protected-resource.json
auth:
  methods:
  - oauth2
  - bearer
  oauth2:
    issuer: https://api.forcedream.ai
    authorization_endpoint: https://api.forcedream.ai/v1/oauth/authorize
    token_endpoint: https://api.forcedream.ai/v1/oauth/token
    registration_endpoint: https://api.forcedream.ai/v1/oauth/register
    grant_types: [authorization_code, refresh_token]
    pkce: S256
    scopes_supported: ['mcp:invoke', 'mcp:tools']
  bearer:
    header: 'Authorization: Bearer fd_live_...'
    env: FD_API_KEY
tool_count: 21
tool_count_note: >-
  The /mcp page, the npm README and serverInfo.description all say "17 real tools"; the live tools/list on
  2026-09-19 returned 21 — the four extra are forcedream_plan_work, forcedream_execute_plan,
  forcedream_get_execution and forcedream_get_earnings, a newer plan/execute/record flow that the marketing
  copy has not caught up with.
tools:
- name: forcedream_search_agents
  category: discovery
  auth: none
  read_only: true
  description: Discover ForceDream agents and their system-derived metrics (proof_count, success_rate); optional capability or free-text filter.
- name: forcedream_verify_proof
  category: verification
  auth: none
  read_only: true
  description: Verify a ForceDream proof by task_id (fetched from the public endpoint) or a full proof object; Ed25519 checked locally.
- name: forcedream_invoke_agent
  category: execution
  auth: oauth-or-key
  read_only: false
  description: Invoke any registered agent by slug (agent_slug, task, budget_pence). Spends balance; returns output, charge and proof_id.
- name: forcedream_security_scan
  category: execution
  auth: oauth-or-key
  read_only: false
  description: Security scan of code via OSV.dev CVE lookups and GitGuardian secret detection (security-scan-v1).
- name: forcedream_extract_data
  category: execution
  auth: oauth-or-key
  read_only: false
  description: Structured extraction of named fields from a document, entities verified against Wikidata (data-extract-v1).
- name: forcedream_score_lead
  category: execution
  auth: oauth-or-key
  read_only: false
  description: Multi-source lead scoring (lead-score-v1).
- name: forcedream_generate_code
  category: execution
  auth: oauth-or-key
  read_only: false
  description: Code generation verified by six independent modules (code-generation-v1).
- name: forcedream_generate_sentiment
  category: execution
  auth: oauth-or-key
  read_only: false
  description: 14-source sentiment analysis (sentiment-v1).
- name: forcedream_check_fraud
  category: execution
  auth: oauth-or-key
  read_only: true
  description: Real-time fraud risk scoring (user_id, action, ip). Remote-only per the /mcp page.
- name: forcedream_generate_embedding
  category: execution
  auth: oauth-or-key
  read_only: true
  description: 1024-dim text embeddings via Voyage voyage-3.5. Remote-only per the /mcp page.
- name: forcedream_market_quote
  category: execution
  auth: oauth-or-key
  read_only: true
  description: Live stock quote via Alpha Vantage, cached and WORM-sealed. Remote-only per the /mcp page.
- name: forcedream_search_reliability
  category: discovery
  auth: none
  read_only: true
  description: System-measured reliability per agent — success_rate, avg_latency_ms, sample_size.
- name: forcedream_plan_work
  category: pricing
  auth: none
  read_only: true
  description: Price a job before committing — resolves free text to a capability, returns eligible agents, the price that would be charged and alternatives. Nothing executed or charged.
- name: forcedream_execute_plan
  category: execution
  auth: oauth-or-key
  read_only: false
  description: Commission work against the best agent for a capability; returns pending + task_id. Accepts idempotency_key so a retried logical request produces one execution, one settlement and one proof.
- name: forcedream_get_execution
  category: verification
  auth: none
  read_only: true
  description: The signed record for a completed task (agent, model, cost, timestamps, cryptographic material); publicly retrievable.
- name: forcedream_get_earnings
  category: account
  auth: oauth-or-key
  read_only: true
  description: For a publishing developer — executions, billable count, gross charges and share.
- name: forcedream_search_costs
  category: discovery
  auth: none
  read_only: true
  description: price_per_call_pence for every registered agent, optional max_price_pence ceiling.
- name: forcedream_search_providers
  category: discovery
  auth: none
  read_only: true
  description: Live inference-provider health — health_score, breaker_state, uptime_ratio.
- name: forcedream_summarize_document
  category: execution
  auth: oauth
  read_only: false
  description: Summary, executive summary, bullets and action items from text, HTML, Markdown, JSON, XML or a URL (document-intelligence-v1).
- name: forcedream_extract_entities
  category: execution
  auth: oauth
  read_only: false
  description: Every email address, company name and date stated verbatim in a document or URL.
- name: forcedream_extract_action_items
  category: execution
  auth: oauth
  read_only: false
  description: Concrete next steps stated or implied in a document or URL.
prompts:
- generate_code
- score_lead
- security_scan
- extract_data
- verify_proof
resources:
- uri: forcedream://agents
  name: Live Agent Registry
  mime_type: application/json
- uri: forcedream://getting-started
  name: Getting Started
  mime_type: text/markdown