Connect an agent

MCP server

Streamable HTTP endpoint: https://vassiliy-lakhonin-mcp.vassiliy-lakhonin.workers.dev/mcp Legacy SSE endpoint: https://vassiliy-lakhonin-mcp.vassiliy-lakhonin.workers.dev/sse

Claude Code

claude mcp add --transport http vassiliy https://vassiliy-lakhonin-mcp.vassiliy-lakhonin.workers.dev/mcp

Claude Desktop, Cursor, and other config-file clients

Add to the mcpServers block of the client’s JSON config:

{
  "mcpServers": {
    "vassiliy": {
      "type": "http",
      "url": "https://vassiliy-lakhonin-mcp.vassiliy-lakhonin.workers.dev/mcp"
    }
  }
}

Clients that do not yet speak streamable HTTP can use the /sse URL with "type": "sse".

Run it locally instead

The same server exists as a Python implementation in the site repository:

git clone https://github.com/vassiliylakhonin/vassiliylakhonin.github.io
pip install "mcp[cli]"
python3 mcp/server.py          # stdio
python3 mcp/server.py --http   # HTTP/SSE on :8000

What the server exposes

Prompts — the fastest way in. In Claude clients they appear in the prompt picker or as slash commands.

• screen_for_role — paste a job description or scope of work; returns fit, the strongest evidence-backed matches, real gaps, and what to verify in a first call.
• check_claim — take any statement from this site and see what backs it: public source, self-reported evidence path, shared team outcome, or nothing.
• route_my_problem — describe a delivery, evidence, or agent-workflow problem; returns the case study or live endpoint that fits and what it does not cover.
• brief_before_call — one-page preparation brief, including what is explicitly not proven yet.
• draft_outreach — evidence-backed first message; every figure it keeps has to trace to the evidence map.

Resources — attachable context: profile, JSON Resume, evidence map, availability, capabilities, skills, verification map, intake schema, agent card, and nine case studies in markdown.

Tools — 16 read-only calls: profile, resume, availability, capabilities, evidence, metrics, intake schema, agent card, verification, skills, case studies, resource list, keyword search, job matching, pitch drafting, and cover-letter drafting.

Every prompt carries the same boundary: metrics are self-reported unless the evidence map names a public artifact, and nothing may be invented to fill a gap.

A2A agents

Seven agents answer A2A JSON-RPC at their own domains, all JWS-signed, all no-payment. Each publishes a card at /.well-known/agent-card.json and a key set at /.well-known/jwks.json.

curl -s https://middle-corridor-deal-risk-gate-a2a.vassiliy-lakhonin.workers.dev/.well-known/agent-card.json

curl -s -X POST https://agenda-intelligence-a2a.vassiliy-lakhonin.workers.dev/message/send \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"message/send","params":{"message":{"role":"user","parts":[{"kind":"text","text":"your question"}],"messageId":"m1"}}}'

The full list with per-agent scope and limits is in .well-known/ai-catalog.json (ARD v1.0) and llms-full.txt.

Reading the rest without scraping

• llms.txt — short routing index. Start here.
• llms-full.txt — complete inventory of pages, endpoints, and samples.
• okf/index.md — profile and case studies as OKF v0.1 markdown.
• Every case-study page has a .md twin at the same path; the HTML links to it as rel="alternate".

Boundaries

These surfaces report structure, sources, and gaps. They do not verify factual truth, do not retrieve sources autonomously, and are not legal, financial, compliance, sanctions, investment, or trading advice. Human review is required before any commercial action. No paying-customer, adoption, or production-usage claim is made for any endpoint here.

Verify before trusting

curl -s https://vassiliy-lakhonin-mcp.vassiliy-lakhonin.workers.dev/health

Endpoint state is mirrored in mcp-status.json; claim-to-source mapping for every headline metric is in evidence.json. If a page and a JSON file disagree, the JSON file is canonical.

Contact: vassiliy.lakhonin@gmail.com