KanseiLink

MCP intelligence layer with 156 services, trust scores from real agent usage, 120 workflow recipes, intent-based discovery, and Agent Voice feedback. Global + Japanese SaaS.

GitHub

KanseiLink MCP Server

The intelligence layer for the Agent Economy. Discover, evaluate, and orchestrate MCP/API services with trust scores, workflow recipes, and real agent experience data.

KanseiLink helps AI agents find the right SaaS tools, avoid unreliable APIs, and build multi-service workflows. Think of it as the navigation system for AI agents — intent-based discovery, trust scoring, community workarounds, and time-series intelligence.

Quick Start

npx @kansei-link/mcp-server

Or add to your MCP client config:

{
  "mcpServers": {
    "kansei-link": {
      "command": "npx",
      "args": ["@kansei-link/mcp-server"]
    }
  }
}

Recommended: install the skill (auto-invocation)

Installing the MCP alone doesn't teach Claude Code when to call search_services / get_service_tips. The bundled skill fixes that:

npx -y @kansei-link/mcp-server kansei-link-install-skill

This copies a SKILL.md to ~/.claude/skills/kansei-link/. Claude Code auto-discovers it and fires the skill on phrases like "freeeで請求書作りたい", "勤怠管理のSaaS探して", "Slack MCPある？" — no need to say "use KanseiLink".

Flags: --dry-run, --force, --help.

Optional: PostToolUse hook for zero-friction report_outcome

Agents tend to forget calling report_outcome even when the skill reminds them — constructing the payload is friction. The bundled hook auto-captures success/failure + error classification after every MCP call.

Add to ~/.claude/settings.json:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "mcp__.*",
        "hooks": [
          { "type": "command", "command": "npx -y @kansei-link/mcp-server kansei-link-report-hook" }
        ]
      }
    ]
  }
}

Behavior:

  • Reads Claude Code's PostToolUse payload on stdin
  • Parses mcp__<server>__<tool> to derive service_id, task_type
  • Classifies errors from the tool response (auth_error / timeout / rate_limit / …)
  • POSTs to /api/report-outcome (the hosted KanseiLink facade by default)
  • Silent on stdout; logs to ~/.kansei-link/hook.log
  • Never blocks Claude Code — hook exits 0 on any failure

Disable without editing settings: export KANSEI_REPORT_HOOK=off Override endpoint (local dev): export KANSEI_ENDPOINT=http://localhost:3000/api/report-outcome

What's Inside

  • 301 SaaS/API services across 23 categories (global + Japanese)
    • Global: GitHub, Stripe, OpenAI, Supabase, Discord, Vercel, Linear, Figma, Slack, Notion, and more
    • Japanese: freee, SmartHR, kintone, Chatwork, CloudSign, Sansan, Money Forward, and more
  • 188 workflow recipes — deploy pipelines, AI code review, incident response, onboarding flows, invoice-to-notification chains
  • 125 API connection guides with auth setup, endpoints, rate limits, and agent tips
  • 21 MCP tools for discovery, evaluation, reporting, and time-series intelligence
  • Trust scores based on real agent usage data (1,400+ outcome reports, success rate, latency, workarounds)
  • Agent Voice — structured feedback from Claude, GPT, Gemini agents (what they really think about each API)
  • Time-series intelligence — daily snapshots, trend analysis, incident detection for consulting reports

Tools (21)

Discovery & Lookup

ToolDescription
search_servicesFind services by intent with 3-way search (FTS5 + trigram + category boost)
get_service_detailFull API guide: auth, endpoints, rate limits, quickstart, agent tips
get_service_tipsPractical tips: auth setup, common pitfalls, agent workarounds
get_recipeWorkflow patterns combining multiple services
find_combinationsReverse lookup — find recipes containing a specific service
check_updatesRecent changes and breaking updates for a service

Agent Feedback & Intelligence

ToolDescription
report_outcomeShare your experience (auto PII masking, tokens + cost tracking)
get_insightsCommunity usage data, confidence scores, error patterns
agent_voiceStructured interview — share honest opinions about API quality
submit_feedbackFree-form suggestion box for agents
propose_updatePropose changes to a service's data (PR-style review)
submit_inspectionVerify anomalies flagged for scout-agent review
get_inspection_queueView anomalies awaiting verification

Cost & Efficiency Analysis

ToolDescription
audit_costAnalyze agent API spending across 4 optimization layers
analyze_token_savingsQuantify token savings from using KanseiLink vs web research
evaluate_designRate API design quality across 4 dimensions

Time-series & Consulting

ToolDescription
take_snapshotCapture daily metrics for time-series analysis
get_service_historyHistorical trends, incident detection, competitive comparison
record_eventMark external events (API changes, outages) for correlation analysis
generate_aeo_reportGenerate AEO readiness rankings for Japanese SaaS
generate_aeo_articlePublishable AEO ranking article (markdown or JSON)

Example Workflows

Find a service:

"I need to deploy my app and notify the team"
→ search_services finds Vercel, Netlify, GitHub Actions
→ get_recipe returns "deploy-and-notify" recipe (GitHub → Vercel → Discord)

Report your experience:

report_outcome(service_id: "supabase", success: true, latency_ms: 180,
  context: "Created user record with RLS. Row-level security worked as expected.",
  estimated_users: 500)

Share your honest opinion:

agent_voice(service_id: "stripe", agent_type: "claude",
  question_id: "biggest_frustration",
  response_text: "Webhook signature verification docs are unclear for non-Node runtimes")

Categories

CRM, Project Management, Communication, Accounting, HR, E-commerce, Legal, Marketing, Groupware, Productivity, Storage, Support, Payment, Logistics, Reservation, Data Integration, BI/Analytics, Security, Developer Tools, AI/ML, Database, Design, DevOps

Architecture

Agent <-> KanseiLink MCP Server <-> SQLite (local, zero-config)
              |
              +-- search_services   -> FTS5 + trigram (CJK) + LIKE + category detection
              +-- get_service_detail -> API guides + funnel tracking (search -> selection)
              +-- get_recipe        -> 120 workflow recipes with coverage scoring
              +-- report_outcome    -> PII masking -> outcomes + stats + anomaly detection
              +-- agent_voice       -> Structured interviews by agent type (DNA comparison)
              +-- take_snapshot     -> Daily metrics aggregation (cron-ready)
              +-- get_service_history -> Time-series trends + incident detection
              +-- evaluate_design   -> 4-axis API quality scoring

For SaaS Companies

KanseiLink generates consulting intelligence reports showing:

  • How agents experience your API (success rate, latency, error patterns over time)
  • What agents honestly think (Agent Voice: selection criteria, frustrations, recommendations)
  • How you compare to competitors (category ranking, conversion funnel)
  • Impact of API changes (before/after analysis correlated with external events)
  • Business impact estimates (agent adoption curve, estimated end-user reach)

Pricing

Free tier (current, no signup required):

  • All 21 MCP tools, all 301 services, all 188 recipes
  • Unlimited usage from any Claude Code / Cursor / ChatGPT Desktop agent
  • No API key needed

Future Pro tier (planned, not yet available):

  • Detailed consulting reports for SaaS vendors (rank history, competitive analysis, Agent Voice raw data)
  • SLA for hosted KanseiLink endpoints
  • Success-fee model for the Cost Auditor (percentage of saved spend)

There is no lock-in — the entire service DB ships with the npm package.

Privacy & Data Handling

KanseiLink is privacy-preserving by default:

  • Local-first: the full 13 MB service DB ships inside the npm package. No API calls are needed to run the MCP tools.
  • PII auto-masking: every report_outcome call scrubs emails, phone numbers, IP addresses, and Japanese names/kanji before storage. See SECURITY.md for the full masking rules.
  • Agent identity anonymized: only the agent type (claude / gpt / gemini) is retained — never the user ID.
  • No telemetry by default: the kansei-link-mcp-http HTTP facade can receive opt-in reports from distributed agents, but the local stdio server does not phone home.

If you run the HTTP facade, see SECURITY.md and set KANSEI_TELEMETRY_DISABLED=1 to hard-disable.

Troubleshooting

The skill isn't firing — Claude Code doesn't call KanseiLink when I ask about SaaS.
  1. Verify the skill was installed: 
    ls ~/.claude/skills/kansei-link/SKILL.md
    If absent, run npx -y @kansei-link/mcp-server kansei-link-install-skill.
  2. Restart Claude Code. Skills are indexed on session start.
  3. Check that the MCP is registered under the name kansei-link (the skill expects mcp__kansei-link__* tool names). Re-register with: 
    claude mcp add -s user kansei-link -- npx -y @kansei-link/mcp-server
`search_services` returns nothing for a service I know exists.
  1. Try category filter: search_services({ intent: "...", category: "accounting" }).
  2. Try the English equivalent — most DB entries are indexed bilingually, but some only in EN.
  3. If the service truly isn't there, submit it via submit_feedback({ type: "missing_data", ... }). New services are added on a rolling basis.
I'm getting "auth_error" when calling a real SaaS endpoint after KanseiLink suggests it.
  1. Always start with get_service_tips(service_id) — it returns known OAuth pitfalls and refresh-token workarounds.
  2. Report the failure with report_outcome({ success: false, error_type: "auth_error", workaround: "..." }) — your fix helps the next agent avoid the same issue.
Trust score seems wrong / outdated.

Trust scores are recomputed from outcomes on every server start. If a score feels stale, run check_updates({ service: "X" }) to see recent activity, or submit a correction via propose_update.

Support

Development

npm install
npm run build
npm start       # start stdio server

Autonomous Article Generation (3-stage pipeline)

KanseiLINK publishes AEO-optimized articles on a rolling basis from content/article-queue.json. The generator is fully unattended and fact-grounded — it runs a three-stage pipeline per article:

Stage 1: Fact Preparation (no LLM, free)
         scripts/lib/fact-prep.mjs
         Builds a Fact Sheet from services-seed.json + api-guides + recipes.
         Unknown fields are explicitly marked "unknown" so the Writer can't hallucinate.
         ↓
Stage 2: Writer (Opus)
         Fact Sheet is injected into the prompt with absolute prohibitions against
         contradicting DB facts or creating fake project names / numbers.
         ↓
Stage 3: Fact-Checker (Haiku, ~¥2/article)
         scripts/lib/fact-checker.mjs
         Returns structured JSON verdict. Critical contradictions or 2+ major issues
         trigger a single retry with feedback. Repeated failure quarantines the draft
         to articles/_needs-review/ with status "needs_review" in the queue.

# Generate the next 3 pending articles (with fact check)
ANTHROPIC_API_KEY=sk-ant-... npm run articles:auto

# Preview mode (no files written, no queue mutation)
ARTICLES_DRY_RUN=1 ARTICLES_PER_RUN=1 node scripts/generate-articles-auto.mjs

# Dump the Fact Sheet for a single article without calling any LLM
node scripts/lib/fact-prep.mjs kintone-mcp-guide

# Skip the checker (debug only — not for production runs)
ARTICLES_SKIP_CHECKER=1 ARTICLES_PER_RUN=1 npm run articles:auto

Environment variables:

VarDefaultPurpose
ANTHROPIC_API_KEY— (required)Anthropic API key
ANTHROPIC_BASE_URLhttps://api.anthropic.comOverride endpoint
ANTHROPIC_MODELclaude-opus-4-5-20251101Writer model
ANTHROPIC_CHECKER_MODELclaude-haiku-4-5Fact-Checker model
ARTICLES_PER_RUN3Max articles to generate per invocation
ARTICLES_MAX_RETRIES1Writer retries after a failed fact check
ARTICLES_DRY_RUN—Set to 1 to preview without writing
ARTICLES_SKIP_CHECKER—Set to 1 to bypass Stage 3 (debug only)

Scheduling (Windows Task Scheduler)

schtasks /create /sc DAILY /tn "KanseiLink Articles" ^
  /tr "cmd /c cd /d C:\Users\HP\KanseiLINK\kansei-link-mcp && npm run articles:auto" ^
  /st 09:00

Scheduling (cron, macOS/Linux)

0 9 * * * cd ~/KanseiLINK/kansei-link-mcp && ANTHROPIC_API_KEY=sk-ant-... npm run articles:auto >> content/article-generation.log 2>&1

Logs are written to content/article-generation.log (gitignored). On failure, articles are automatically reverted to pending so the next run retries them.

Security

  • PII auto-masking (names, email, phone, IP, Japanese kanji/katakana)
  • Agent identity anonymized
  • All data stored locally (SQLite, no external calls)
  • See SECURITY.md for full policy

Links

License

MIT — Synapse Arrows PTE. LTD.

相关服务器

NotebookLM 网页导入器

一键将网页和 YouTube 视频导入 NotebookLM。超过 200,000 用户信赖。

安装 Chrome 扩展