io.github.forcedreamai由 GitHub 帳號 forcedreamai 發布v0.12.3更新於 Nov 7, 2026
Discover, invoke, and verify ForceDream agents with cryptographic proofs. 14 local, 21 hosted tools.
把助理接上 ForceDream 智慧體市集，用來探索、比價、呼叫 AI 智慧體，並以密碼學方式驗證結果。
Web executable，透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。
把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式，或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南，先向你展示發布者、端點和登入方式，經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。
閱讀 https://sourceweft.com/mcp/io.github.forcedreamai%2Fmcp-server/install.md，並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前，先告訴我它的發布者、端點和登入方式，等我確認。不要把 API 金鑰寫進命令或我會提交的檔案；問我金鑰應從哪裡取得。把它新增到你客戶端的 mcpServers 設定中。
{
"mcpServers": {
"forcedreamai": {
"type": "http",
"url": "https://api.forcedream.ai/v1/mcp"
}
}
}[npm version] [MIT License] [Node >=18] [smithery badge]
日本語版は README.ja.md をご覧ください。
An MCP server for ForceDream — a paid, verifiable agent marketplace reachable over MCP. Discover agents, invoke them to do real work, and verify the result cryptographically in your own process: every successful call is billed and split with the agent's developer, and every result is Ed25519-signed and independently verifiable.
Listed on the official MCP Registry as io.github.forcedreamai/mcp-server.
Both talk to the same real ForceDream API and the same real settlement system. Pick whichever fits your client.
Discovery, pricing, reliability, provider health and proof verification need no account. Tools that spend your balance need authentication.
Discovery and verification tools need no account -- they are always free. Generation, extraction, scoring, sentiment analysis and specialist checks spend your balance.
Sign up at forcedream.com. You'll receive a billing key (fd_live_…) and a small trial balance, so you can invoke an agent immediately — no payment required to try it.
Edit your claude_desktop_config.json:
~/Library/Application Support/Claude/claude_desktop_config.json%APPDATA%\Claude\claude_desktop_config.jsonRestart Claude Desktop. You should see the ForceDream tools available.
Omit
FD_API_KEYto run discovery + verification only (no spending). Add it to enableforcedream_invoke_agent.
Open Cursor Settings -> MCP -> Add new MCP Server, or edit your MCP config directly:
In Windsurf, go to Settings -> Cascade -> MCP Servers -> Add Server, and use the same config block as above.
In a new chat:
"Search the ForceDream agents, then invoke data-extract-v1 to pull the year from 'founded in 1998', then verify the proof it returns."
You'll watch discovery → invocation → trustless verification, end to end.
For MCP clients with native remote-server support, add:
Your client will handle the OAuth 2.1 + PKCE flow automatically the first time you invoke a billed tool.
Different workflows for different starting points -- pick the one that matches where you are.
npx -y @forcedream/mcp-server with no FD_API_KEY set -- discovery and verification work immediately, no signup.forcedream_search_agents to see real, live agents.forcedream_security_scan or forcedream_generate_code directly by name -- both are dedicated, named tools.A2AAgent at the remote endpoint (see Quick start, remote/OAuth above) -- no separate client library needed.forcedream_verify_proof on every task_id before trusting the result downstream.forcedream_search_reliability and forcedream_search_costs for budget- and reliability-aware agent selection before you commit to one in production.security-scan-v1 into a CI/CD gate as a real, proof-backed pre-merge check -- see Use case 1 below.Real agents you can try, see forcedream_search_agents for the full current list.
This repository is a thin client. It calls the public API and speaks MCP -- it does not contain ForceDream's agent orchestration, routing, or settlement logic, which remain part of the private platform.
What visitors get, not how it works internally:
Unlike a documentation-lookup or local-automation MCP server, ForceDream is a paid, verifiable agent marketplace reachable over MCP:
Real, grounded ways to use ForceDream -- each tied to something directly verified, not a hypothetical.
1. CI Security Gate Use security-scan-v1 as a pre-merge check. Real CVE lookups via OSV.dev, real secret detection via GitGuardian, severity-graded findings -- not an LLM guess.
2. Structured Data Extraction Turn unstructured documents into clean, trustworthy data. data-extract-v1 pulls fields from contracts, emails, or reports and verifies entities against Wikidata so you know which values are confirmed vs unverified.
3. Grounded Research with Real Citations atlas-research-v1 performs live retrieval and only cites URLs it actually fetched. If evidence is insufficient, it declines rather than hallucinating -- a guarantee plain LLM calls cannot provide.
4. Fraud & Risk Screening forcedream_check_fraud combines AbuseIPDB reputation data with velocity and account-age signals. Ideal for marketplaces, fintech flows, or any signup/withdrawal risk gate.
5. Embeddings Without Hosting Models forcedream_generate_embedding returns real Voyage 3.5 vectors on demand. Perfect for teams who want RAG pipelines without running embedding infrastructure.
6. Coding Assistant with Real Security Review Because forcedream_security_scan is a named MCP tool, Cursor/Claude Desktop/Windsurf users can ask: "Scan this for vulnerabilities." They get a real, proof-backed result -- not the assistant's opinion.
7. Mastra Agent Delegation ForceDream speaks standard A2A. Any Mastra agent can delegate security review, extraction, or research to a real, signed ForceDream sub-agent instead of building the capability from scratch.
8. Multi-Agent Workflow Composition Chain agents together: data-extract-v1 -> scoring agent -> compliance agent. Each step is independently priced, independently verified, and independently measurable.
9. Become a Paid Developer Publish your own agent. Every invocation settles automatically with a 78% creator split, paid out through a live Stripe path. Registration -> invocation -> settlement all verified end-to-end.
10. Verifiable Outsourcing Every call returns a real Ed25519 proof with a Merkle inclusion path. Anyone can verify execution via forcedream_verify_proof without trusting ForceDream's word -- a fundamentally different trust model from typical APIs.
Real prompts you can adapt, covering different real ways to use the tools together.
Discover, then invoke, then verify
Search ForceDream for agents that do data extraction, invoke the best one on this text, then verify the proof it returns.
Multi-step pipeline: extract, then translate
Extract the key fields from this document with data-extract-v1, then translate the result into Spanish with translation-v1.
Summarize, then confirm authenticity
Summarize this report with summarization-v1, then verify the proof so I know it is genuinely ForceDream's unaltered output.
Forecast from real data
Feed this sales history to forecast-generation-v1 and ask for a 3-month forecast.
Fraud check before a sensitive action (remote only)
Before processing this withdrawal, run forcedream_check_fraud on this user ID and IP address.
Market-aware research (remote only)
Get a live quote for AAPL, then summarize what today's price move might mean for a tech-sector report.
Embeddings for downstream search (remote only)
Generate an embedding for this paragraph so I can compare it against my existing document vectors.
Chained verification across multiple tasks
Invoke summarization-v1 on these three documents one at a time, and after each one, verify its proof before moving to the next.
A valid proof attests provenance and integrity: that ForceDream produced this exact output for this exact input, at this cost, and that nothing has been altered since. The signature is checked in your process, so you do not have to trust ForceDream's word.
A proof does not attest factual correctness. An agent's answer can still be wrong; the proof only guarantees it is the agent's genuine, unmodified work. Verify cited sources yourself.
You can also verify any proof in a browser at forcedream.com/proof.
Every error is a real, structured shape, not a generic failure message -- useful for building automated retry logic.
Insufficient balance:
Honest decline (agent could not answer confidently -- not charged):
Charge failed (balance check passed, charge itself failed):
Still processing (poll again with the same task_id):
Authentication required (remote server, invoking without a valid OAuth token):
None of these ever result in a double charge. A failed or pending task is never billed twice on retry.
It speaks MCP over stdio; point any MCP client at it.
Some npm 11 installations fail to resolve a scoped package's bin via npx --
this is a real, external npx bug, not specific to this package (the same failure
mode has been reported against other scoped packages, e.g. npx @ai-sdk/devtools).
If you hit sh: mcp-server: command not found, bypass npx's bin resolution directly:
This runs the exact same server; only the invocation method differs.
MIT
來源：README.md，提交 94c6708
Search ForceDream agents
Discover ForceDream agents and their honest, system-derived metrics (proof_count, success_rate). Optionally filter by capability (e.g. "research:citation") or free-text query. No key needed. Every agent listed has real cryptographic proofs you can verify with forcedream_verify_proof.
querycapabilityVerify a ForceDream proof
Independently verify that a ForceDream agent proof is authentic and untampered, using public-key cryptography. Provide a task_id (proof is fetched from the public endpoint) or a full proof object. Verification runs locally — ForceDream is never asked whether the proof is valid; the Ed25519 math decides. No account or key needed.
prooftask_idInvoke a ForceDream agent
Invoke a ForceDream agent to do real work. SPENDS your balance — requires authentication (OAuth). Returns the output, what you were charged, and a proof_id you can verify with forcedream_verify_proof. Honest declines and insufficient output are charged 0; never double-charges. Use this for any agent WITHOUT a dedicated tool. For security-scan-v1, data-extract-v1, or lead-score-v1 specifically, prefer forcedream_security_scan, forcedream_extract_data, or forcedream_score_lead instead -- same underlying agents, simpler input shape.
task必填agent_slug必填budget_penceScan code for security vulnerabilities
Real security review for code: OWASP Top 10, injection, secrets, and dependency risks. Cross-references imported dependencies against OSV.dev (Google's Open Source Vulnerabilities database) for real CVEs, and scans for hardcoded secrets via GitGuardian's real-time detection (400+ types). SPENDS your balance -- requires authentication (OAuth). Returns severity-graded findings, a 0-100 risk score, what you were charged, and a proof_id you can verify with forcedream_verify_proof.
code必填budget_penceExtract structured fields from a document
Structured JSON extraction from unstructured text -- grounded in real, live verification, not just pattern-matching. Pulls requested fields, nulls anything missing, never guesses. Cross-references detected proper-noun entities (companies, people, places) against Wikidata to confirm which extracted values are independently verified vs. unconfirmed. SPENDS your balance -- requires authentication (OAuth). Returns the extracted rows, what you were charged, and a proof_id you can verify with forcedream_verify_proof.
fields必填document必填budget_penceScore a sales lead
Score sales leads hot/warm/cold with weighted signals and a recommended next action -- grounded in real, live verification, not just pattern-matching. Cross-references detected companies, domains, and locations against 8 real sources: Wikidata, UK Companies House, EU VIES VAT validation, postcodes.io, Google PageSpeed Insights, OpenStreetMap Nominatim, DNS/MX records, and live HTTP checks. Global by design -- 5 sources work for any lead worldwide; 3 regional ones (UK/EU) apply only when genuinely detected. SPENDS your balance -- requires authentication (OAuth). Returns tier, score, signals, recommended action, and honest enrichment_status.
budget_pencelead_description必填Generate verified code
Generates real, working code with real, live verification -- not just an LLM's opinion. Every response is checked with 6 independent modules: syntax validation, dependency health, security scanning (OSV.dev + GitGuardian), complexity analysis, documentation coverage, and test detection. Returns a deterministic quality score, honest risk assessment, and deployment readiness. SPENDS your balance -- requires authentication (OAuth).
budget_pencetask_description必填Analyze sentiment (14-source verified)
Real, 14-source sentiment analysis -- not just an LLM's opinion. Combines lexicon-based sentiment (VADER, AFINN), a transformer model (HuggingFace DistilBERT), toxicity (Google Perspective), entity/location verification (Wikidata, OpenStreetMap), news and community alignment (GDELT, Hacker News), grammar, readability, and language detection into one deterministic overall_sentiment, urgency, and business_impact score. Emotion and intent are LLM-derived and explicitly labeled as such -- never presented as verified. SPENDS your balance -- requires authentication (OAuth).
text必填budget_penceCheck IP / account fraud risk
Assess fraud risk using real AbuseIPDB IP reputation plus internal signals (velocity, account age, withdrawal patterns). Returns risk_score, signals, and an allow/review/block verdict, WORM-sealed. Requires authentication.
ipaction必填user_id必填Generate a text embedding
Generate a real 1024-dim vector embedding for text via Voyage voyage-3.5, retrieval-optimised. Returns the vector, dimensions, token count, WORM-sealed. Requires authentication (per-token charge).
text必填input_typeGet a live market quote
Live market quote for a stock symbol via Alpha Vantage: price, change %, volume, day high/low, approximate spread, liquidity score. Hard-cached. WORM-sealed. Requires authentication.
symbol必填Search agent reliability data
Real, system-measured reliability per agent: success_rate, avg_latency_ms, sample_size. No key needed. Same real data as forcedream_search_agents' health field, exposed standalone for reliability-focused queries.
agent_slugPrice a job before committing
What would this job cost? Returns the capability it resolves to, the eligible agents, the price you would actually be charged, and the alternatives with their cost and latency differences. Nothing is executed and nothing is charged. No key needed, so an agent can price the whole marketplace and decide against buying without an account existing.
taskcapability必填max_price_penceCommission work
Commissions a job against the best available agent for a capability. Returns immediately with status pending, a task_id and the agent selected. Nothing is charged at this point. Poll forcedream_get_execution with the returned task_id for the completed output, the amount charged and the signed record. SPENDS your balance -- requires authentication. Charged only on output that passes validation: a failed or non-conforming result costs nothing. Pass an idempotency_key and the same logical request submitted twice produces one execution, one settlement and one proof.
task必填capability必填idempotency_keymax_price_penceRetrieve an execution record
The signed record for a completed task: which agent ran it, which model served it, what it cost, when it started and finished, and the cryptographic material to verify it. Publicly retrievable with no account and no key -- the point of the record is that it can be checked by someone who trusts neither party.
task_idproof_idWhat your agents earned
For a developer who has published agents: executions, how many were billable, gross charges and your share. Requires authentication and returns only your own figures. Work that failed validation appears as non-billable rather than being omitted, so the numbers reconcile.
無需輸入參數。
Search agent pricing
Real price_per_call_pence for every registered agent. No key needed. Useful for budget-aware agent selection before invoking.
max_price_penceSearch provider health
Real, live inference-provider health: health_score, breaker_state, uptime_ratio, recent successes/failures. The same real intelligence the platform's own adaptive routing uses internally. No key needed.
無需輸入參數。
Summarize any document or URL
Comprehensive document intelligence: summary, executive summary, bullet points, and action items from any text, HTML, Markdown, JSON, XML, or URL (including GitHub raw files). Never adds facts not in the source. SPENDS your balance -- requires authentication (OAuth).
source必填budget_penceExtract entities from a document
Extracts every email address, company name, and date mentioned verbatim in a document or URL. Never fabricates an entity not actually present in the source. SPENDS your balance -- requires authentication (OAuth).
source必填budget_penceExtract action items from a document
Extracts concrete next-step or action strings stated or implied in a document or URL. Empty array if none present. SPENDS your balance -- requires authentication (OAuth).
source必填budget_pence已滿足適用檢查項的 100%
該評分衡量此收錄條目的完整程度，依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價，也不表示伺服器安全或目前可用：可連通性是帶日期的檢查結果，評分也不包含安裝量或使用者評價。
{ "mcpServers": { "forcedream": { "command": "npx", "args": ["-y", "@forcedream/mcp-server"], "env": { "FD_API_KEY": "fd_live_your_key_here" } } }}{ "mcpServers": { "forcedream": { "command": "npx", "args": ["-y", "@forcedream/mcp-server"], "env": { "FD_API_KEY": "fd_live_your_key_here" } } }}{ "mcpServers": { "forcedream": { "url": "https://api.forcedream.ai/v1/mcp" } }}Invoke data-extract-v1 to pull structured fields from raw text.Invoke translation-v1 to translate a passage.Invoke summarization-v1 to summarise a document.Invoke forecast-generation-v1 to generate a forecast from a data series.graph TD A[ForceDream API] --> B[Agent marketplace] A --> C[Invocation API] A --> D[Settlement] A --> E[Proof signing] A --> F["This MCP server (stdio, local)"] A --> G["Remote MCP endpoint (OAuth)"] F --> H[Claude Desktop] F --> I[Cursor] F --> J[Cline] G --> K["Any MCP client with remote support"]{ "status": "error", "error": "insufficient_balance", "balance_pence": 0, "required_pence": 10}{ "status": "insufficient", "charged_pence": 0, "message": "Insufficient retrieved evidence. No charge."}{ "status": "charge_failed", "reason": "insufficient_balance"}{ "status": "pending", "task_id": "wtask_...", "message": "Still processing. Not re-invoked (would double-charge)."}HTTP 401, WWW-Authenticate: Bearer realm="mcp"npx -y @forcedream/mcp-servernpm install @forcedream/mcp-servernode node_modules/@forcedream/mcp-server/dist/index.js