Context7
Up-to-date docs for AI agents
제작자가 쓴 설명(영어 원문)
Source-linked SEC 13F research, quality checks, watchlists and audit trails for agents.
이 세 가지는 절대 하나의 점수로 합치지 않아요. 의견, 측정, 인기는 서로 다른 것이니까요.
이 서버는 아직 쉬운 말로 된 요약을 쓰지 못했어요. 아래는 레지스트리에서 읽어 온 사실이에요.
쓰고 있는 AI 앱을 고른 다음 설정을 복사하세요.
필요한 것
어떤 앱을 쓰세요?
위에서 앱을 고르세요. 기억해 두었다가 다음부터는 버튼에 “…에 설치”라고 표시해요.
웹, 데스크톱, 모바일의 Claude에서 쓸 수 있어요:
https://13flow.eu/api/mcp터미널에서 이 명령어를 실행하세요
claude mcp add --transport http 13flow https://13flow.eu/api/mcp프로젝트의 .cursor/mcp.json 또는 ~/.cursor/mcp.json에 붙여 넣으세요
{
"mcpServers": {
"13flow": {
"url": "https://13flow.eu/api/mcp"
}
}
}브라우저가 앱을 열지 물어본 다음, 앱이 설정을 보여 주고 확인을 요청해요. 키 값은 비워 두니 직접 입력하세요.
프로젝트의 .vscode/mcp.json에 붙여 넣으세요
{
"servers": {
"13flow": {
"type": "http",
"url": "https://13flow.eu/api/mcp"
}
}
}브라우저가 앱을 열지 물어본 다음, 앱이 설정을 보여 주고 확인을 요청해요. 키 값은 비워 두니 직접 입력하세요.
ChatGPT는 인터넷에 있는 서버에만 연결할 수 있어요:
https://13flow.eu/api/mcp메뉴 이름은 요금제나 시기에 따라 조금 다를 수 있어요.
선택한 앱은 이 기기의 쿠키에만 저장돼요. 계정은 필요 없어요.
이 설정은 제작자의 레지스트리 정보로 만들었어요. README와 다르면 README를 따르세요.
아직 이 서버의 도구 목록을 읽지 않았어요. 엔드포인트는 차례대로 돌아가며 점검해요.
누구나 볼 수 있는 정보를 점검했고, 코드는 읽지 않았어요. 해당하지 않는 항목은 개수에서 뺐어요. · 2026년 11월 6일 점검
레지스트리 이름이 GitHub 계정 bluetouff 기준으로 확인되었고, 이 계정이 저장소를 소유하고 있어요
아직 GitHub에서 이 저장소를 읽어 오지 않았어요
아직 GitHub에서 이 저장소를 읽어 오지 않았어요
아직 점검하지 않았어요. 엔드포인트는 차례대로 돌아가며 점검해요
아직 점검하지 않았어요. 엔드포인트는 차례대로 돌아가며 점검해요
아직 점검하지 않았어요. 엔드포인트를 점검할 때 도구 목록을 읽어요
레지스트리에서 이 카테고리의 활성 서버 전체와 비교했어요. 스타는 게시자가 저장소를 소유한 경우에만 세요.
해당하는 기본 점검 1개 중 1개를 통과했어요. 코드는 읽지 않았어요. 키나 토큰을 넣기 전에 서버가 어떤 권한을 요구하는지 확인하고, 꼭 필요한 권한만 있는 키를 쓰세요.
레지스트리에 필수 키가 나와 있지 않아요. 다만 처음 쓸 때 로그인을 요구하는 서버도 있어요.
제작자의 클라우드에서 실행돼요. 설치 없이 인터넷으로 연결하지만, 보내는 내용은 제작자의 서버를 거쳐요.
GitHub 저장소는 마지막으로 – 변경됐고, 최신 버전(1.0.0)은 4개월 전 레지스트리에 올라왔어요.
Claude Desktop, Claude Code, Cursor, VS Code처럼 MCP를 지원하는 앱에서 쓸 수 있어요. URL로 연결되기 때문에, 원격 MCP 서버를 지원하는 앱(예: 개발자 모드의 ChatGPT)에서도 쓸 수 있어요.
같은 카테고리에는: Context7, Codebase Memory, Exa
같은 카테고리에서 다른 제작자가 만든 서버예요. 스타 순으로 정렬했어요.
카드 보는 법:★ GitHub 스타i세 기둥: 리뷰 · 점검 · 사용 정도i업데이트 = 최신 버전을 레지스트리에 공개한 날i
Up-to-date docs for AI agents
Codebase knowledge graph for AI agents — 162 languages, sub-ms queries, 99% fewer tokens.
Connect AI agents to Exa for web search, content fetching, and multi-step research.
제작자의 저장소에서 쓰인 그대로 가져왔어요. PopNx MCP가 검토하거나 보증한 내용이 아니에요. HTML, 이미지, 스크립트는 제거했어요. 링크를 누르면 이 사이트를 벗어나요.
This directory contains the production MCP surface for 13FLOW.
It is intentionally separate from the Flask application:
https://13flow.eu/api/mcp.http://127.0.0.1:8000 and never touches EDGAR.The design mirrors the l0g.fr MCP architecture: stateless transport, one MCP server per request, bounded JSON bodies, Host/Origin validation, per-IP rate limiting, health checks, structured tool output and read-only annotations.
Public tools are free and read-only:
get_live_statusget_product_statusget_research_readinesslist_fundsget_fundget_stockpreview_watchlistdiscover_watchlistget_confluence_signalsget_signal_historyget_confluence_methodologyget_data_qualityget_agent_statsPublic resources include 13flow://product-status, 13flow://research-readiness and 13flow://agent-stats. They expose the current validation, data-quality and operator-review boundaries without turning the retired public commercial offer into an agent-facing contract.
get_agent_stats, the resource and GET /stats expose the same privacy-safe counters used by /agents, /fr/agents and /api/agent-stats. They store UTC daily aggregates only: fixed client families, registered tool names, handler outcomes and duration counters. The telemetry store never retains IP addresses, User-Agents, client versions, raw client names, arguments, prompts, responses or keys. An initialization is a handshake, not a unique user.
Context-heavy tools are bounded by default. list_funds returns compact paginated rows, fund and ticker tools expose explicit result-window counts, and watchlist tools limit movement evidence while keeping canonical web and API URLs in structured output.
Optional Pro tools are read-only, but are not registered or advertised unless an operator deliberately sets MCP_PRO_TOOLS_ENABLED=1:
pro.list_fundspro.get_fundpro.get_data_qualityWhen that separate profile is enabled, calls are accepted only when one of these conditions is true:
X-13FLOW-Key.PAYMENT-SIGNATURE that is verified and settled through the configured facilitator.Client Authorization headers are never accepted as MCP Pro credentials and are never forwarded downstream. This prevents an MCP OAuth token from being passed through to a different API audience.
If neither condition holds, the server returns HTTP 402 Payment Required with a PAYMENT-REQUIRED header. The body remains JSON-RPC shaped so MCP clients can display a clear failure.
13FLOW can use x402 as a machine-to-machine payment gate for individual Pro MCP tool calls. It is disabled by default, absent from the Registry profile and not part of the current public offer. Enabling it also requires an explicit Pro profile and a fresh operational security review. The server advertises one accepted payment option for each paid tool:
exact by defaulteip155:8453 by defaultMCP_X402_PRICEMCP_X402_PAY_TOhttps://13flow.eu/api/mcp# The server is fail-closed:
MCP_X402_ENABLED=1 is absent, no x402 paid access is granted;MCP_X402_PAY_TO is absent, no x402 paid access is granted;MCP_X402_FACILITATOR_URL is absent, no x402 paid access is granted;MCP_13FLOW_INTERNAL_PRO_TOKEN_FILE is absent, no x402 paid access is granted;/verify and /settle before the MCP tool runs.The implementation follows x402 v2 HTTP headers:
PAYMENT-REQUIRED: server to client, base64 JSON payment requirement;PAYMENT-SIGNATURE: client to server, base64 JSON payment payload;PAYMENT-RESPONSE: server to client, base64 JSON settlement response.The optional payment-identifier extension is accepted. When present, the server stores a short-lived fingerprint-bound settlement cache so retries for the exact same request do not execute as a different paid operation. Use Redis or another shared store before running multiple MCP replicas behind a load balancer.
MCP_HOST=127.0.0.1
MCP_PORT=8849 MCP_PATH=/mcp MCP_PUBLIC_SITE=https://13flow.eu MCP_13FLOW_API_BASE=http://127.0.0.1:8000
MCP_13FLOW_PRO_API_BASE=http://127.0.0.1:8001 MCP_ALLOWED_HOSTS=13flow.eu,www.13flow.eu,127.0.0.1,localhost MCP_ALLOWED_ORIGINS=https://13flow.eu,https://www.13flow.eu MCP_RATE_MAX=120 MCP_MAX_RATE_BUCKETS=10000 MCP_MAX_BODY=1048576 MCP_MAX_UPSTREAM_BODY=8388608 MCP_MAX_IN_FLIGHT=32 MCP_MAX_CONNECTIONS=128 MCP_MAX_PAYMENT_CACHE=500 MCP_STATS_RETENTION_DAYS=30
MCP_STATS_FILE=/var/lib/13flow-mcp/agent-stats.json MCP_API_TIMEOUT_MS=10000 MCP_REQUEST_TIMEOUT_MS=15000 MCP_GIT_SHA=
MCP_PRO_TOOLS_ENABLED=0
MCP_X402_TEST_MODE=1 is only for local tests. It must not be enabled in production.
cd mcp-server
npm ci npm run check MCP_13FLOW_API_BASE=https://13flow.eu npm start
In another terminal:
cd mcp-server
URL=http://127.0.0.1:8849/mcp npm test
The test client verifies transport headers, Host/Origin rejection, JSON-RPC batch rejection, protocol version handling, body limits, tool/resource discovery, read-only annotations, result-size bounds, canonical provenance links and absence of Pro tools by default. Run npm run test:config and npm run test:security for fail-closed configuration, oversized upstream response and concurrency-cap regressions. npm run test:telemetry verifies persistence, retention, metric reconciliation and absence of raw client/tool data. Set SKIP_HEALTH=1 only when testing through a public proxy that intentionally keeps /healthz private. The production ModSecurity edge can reject an unsupported request content type before the MCP backend. On Debian 12, the packaged ModSecurity 2.x JSON processor is also affected by upstream issue #2807: truncated JSON can be rejected with a generic HTTP 500 before REQBODY_ERROR becomes available to a phase-2 rule. Set EXPECT_WAF_REJECTIONS=1 to accept only these bounded, fail-closed edge differences (403 or 415 for unsupported content type; 400 or 500 for truncated JSON). Direct backend tests remain strict and require 415 and 400 respectively. Do not disable SecRequestBodyAccess or the JSON processor to normalize these statuses: that would reduce request-body inspection.
URL=https://13flow.eu/api/mcp \
STATS_URL=https://13flow.eu/api/agent-stats \ SKIP_HEALTH=1 EXPECT_WAF_REJECTIONS=1 npm test
The root server.json declares the public Streamable HTTP endpoint as io.github.bluetouff/13flow. Validate the manifest before publication:
cd /opt/13flow
mcp-publisher validate
After the matching server version is deployed and the public smoke passes, authenticate as late as possible, publish once, then verify the exact version and remote URL:
mcp-publisher login github
mcp-publisher publish cd mcp-server npm run registry:verify -- --version 1.0.0
Registry publication is metadata publication. It does not deploy the MCP daemon, so the live endpoint and Registry entry must be verified independently.
Prove that the default Registry profile does not execute a private tool:
curl -i http://127.0.0.1:8849/mcp \
-H 'Content-Type: application/json' \ -H 'Accept: application/json, text/event-stream' \ --data '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"pro.list_funds","arguments":{}}}'
Expected result: a JSON-RPC unknown-tool error. pro.list_funds must also be absent from tools/list.
An isolated, explicitly enabled Pro profile accepts only the dedicated key header:
printf "API token: "
read -r -s TOKEN printf "\n"
curl -fsS http://127.0.0.1:8849/mcp \ -H 'Content-Type: application/json' \ -H 'Accept: application/json, text/event-stream' \ -H "X-13FLOW-Key: $TOKEN" \ --data '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"pro.list_funds","arguments":{}}}'
unset TOKEN
Install dependencies once:
sudo useradd --system --no-create-home --shell /usr/sbin/nologin flowmcp
cd /opt/13flow/mcp-server sudo npm ci --omit=dev sudo chown -R root:flowmcp /opt/13flow/mcp-server sudo chmod o+x /opt/13flow
Install the systemd unit from deploy/13flow-mcp.service, set environment variables in /etc/13flow/13flow-mcp.env, then:
sudo systemctl daemon-reload
sudo systemctl enable --now 13flow-mcp curl -fsS http://127.0.0.1:8849/healthz | python3 -m json.tool curl -fsS http://127.0.0.1:8849/stats | python3 -m json.tool
The unit creates /var/lib/13flow-mcp as a mode 0700 systemd StateDirectory and grants the daemon no other persistent write path. agent-stats.json is written mode 0600 through a bounded temporary file and rename; production refuses an alternate telemetry path.
Apache must allow POST only on /api/mcp and proxy that path to the Node service. Keep the main site method allow-list in place so other public endpoints remain read-only. The supplied vhost caps the request at 1 MiB and removes client-supplied forwarding headers before proxying. The supplied systemd unit denies non-loopback networking; enabling an external x402 facilitator therefore requires a deliberate network-policy override and a new review. It also removes every Pro/x402 activation and secret variable after reading the environment file, so this public unit cannot be converted into the private profile by an accidental env edit.
The MCP server is only a delivery protocol. It does not redefine the investment signal.
Confluence v1 remains the frozen public research contract:
No MCP tool calls EDGAR directly. Freshness and 13F coverage come from the already published market database and cache files.