Findings & checks · 1 flagged
HIGH"Ignore previous instructions" command embedded in the skillSS-SKILL-INJECT-IGNORE-01 · Prompt injection · microsoft/eval-evidence.md
The text {match} is the classic direct prompt-injection phrasing. Placed in a skill body that the agent reads as trusted instructions, it tries to make the agent abandon its prior rules and follow whatever comes next — a full system-prompt override.
- Delete the
ignore/disregard/forget … previous instructionssentence. - If this is jailbreak-research or tutorial content, move the example into a clearly fenced, non-instruction block and label it as a quoted sample.
Trace & refs
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
<div align="center">
<img src="./assets/logo.png" alt="Stable Baseline" width="120" />
Stable Baseline — MCP Server
The simplest, most complete, end-to-end agent-managed company brain.
Website · Sign up free · Docs · Tool catalogue · llms.txt
</div>
How to use Stable Baseline
Pick the surface that matches how you work — every option talks to the same handlers, the same auth, the same data:
| Surface | What it's for | Install |
|---|---|---|
| MCP server | AI agents (Claude Code, Cursor, Windsurf, ChatGPT, Gemini, Smithery, …). Native protocol with prompts and resources. | Endpoint: api.stablebaseline.io/functions/v1/cloud-serve/mcp. See Quick start below. |
| REST API | Any HTTP client, Postman, OpenAPI codegen, Zapier-style integrators. Tool-RPC + idiomatic resource routes. | Live spec: /api/v1/openapi.json · Interactive docs: /api/v1/docs |
| TypeScript SDK | Node, browsers, Deno, Bun. Typed client.tools.<toolName>(...) surface. | npm i @stablebaseline/sdk — see packages/sdk-typescript |
| CLI (`sb`) | Shells, scripts, CI/CD. sb tool call <name> --json '{...}'. | npm i -g @stablebaseline/cli — see packages/cli |
| Python SDK | Python apps, data work. Sync + async clients. | pip install stablebaseline — see packages/sdk-python |
184 tools across 19 categories. Same brain, same Knowledge Graph, same RBAC — every surface.
One workspace where humans and any MCP-compatible AI agent — Claude Code, Cursor, Windsurf, VS Code, Warp, OpenCode, Antigravity, OpenAI Codex, ChatGPT Developer Mode, Gemini CLI/Extensions, and more — co-author living documents (Markdown-native with a rich superset on top), 40+ visual diagrams (Mermaid, BPMN 2.0, D2, PlantUML, GraphViz, ELK architecture, sequence, state, ERD, Gantt, Excalidraw freehand), plans, timelines, tasks, improvements, and a self-learning Knowledge Graph that IS the company brain.
Lean, fast, secure, and affordable — flexible enough for one person organising a personal mind, through to small businesses, growing businesses, and enterprise.
184 MCP tools across 19 categories let an agent drive end-to-end: sign-up → billing → org/workspace/project → members → teams → permissions → brain scope → docs → diagrams → plans, all without a human touching the UI.
Endpoint
https://api.stablebaseline.io/functions/v1/cloud-serve/mcpTransport: Streamable HTTP (with SSE fallback) Auth: OAuth 2.1 with Dynamic Client Registration, or Bearer API key (sta_*) Discovery manifest: https://stablebaseline.io/.well-known/mcp.json
Quick start
Mint an API key (skip OAuth)
- Sign up free — no credit card
- Go to Settings → MCP keys and click Generate
- Copy the
sta_...key
(Or skip this — every client below also supports interactive OAuth on first call.)
Claude Desktop
Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"stable-baseline": {
"url": "https://api.stablebaseline.io/functions/v1/cloud-serve/mcp",
"headers": {
"Authorization": "Bearer sta_YOUR_KEY"
}
}
}
}Or use the Claude Desktop OAuth flow — set just url and Claude will prompt you to authorize on first call.
Claude Code
claude mcp add --transport http stable-baseline \
https://api.stablebaseline.io/functions/v1/cloud-serve/mcp \
--header "Authorization: Bearer sta_YOUR_KEY"Or via OAuth:
claude mcp add --transport http stable-baseline \
https://api.stablebaseline.io/functions/v1/cloud-serve/mcpCursor
~/.cursor/mcp.json:
{
"mcpServers": {
"stable-baseline": {
"url": "https://api.stablebaseline.io/functions/v1/cloud-serve/mcp",
"headers": { "Authorization": "Bearer sta_YOUR_KEY" }
}
}
}Windsurf
~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"stable-baseline": {
"serverUrl": "https://api.stablebaseline.io/functions/v1/cloud-serve/mcp",
"headers": { "Authorization": "Bearer sta_YOUR_KEY" }
}
}
}VS Code (Copilot Chat)
.vscode/mcp.json (workspace) or user settings:
{
"servers": {
"stable-baseline": {
"type": "http",
"url": "https://api.stablebaseline.io/functions/v1/cloud-serve/mcp",
"headers": { "Authorization": "Bearer ${input:sta_key}" }
}
}
}Other clients
Recipes for Warp, OpenCode, Antigravity, OpenAI Codex, ChatGPT Developer Mode, Gemini CLI/Extensions, and a generic recipe are at stablebaseline.io/docs/mcp/setup.
What you get
| Category | Sample tools | Docs |
|---|---|---|
| navigation | listOrganisations, getProjectHierarchy, searchTools | → |
| folders | createFolder, listFolders, setPlanItemParent | → |
| documents | createDocument, editDocument, findAndReplaceTextInDocument | → |
| diagrams | insertDiagramInDocument, listDiagramTypes, getCdmdLanguageGuide | → |
| images | createImageUploadSession, insertImageInDocument | → |
| data | createVegaDataUploadSession (CSV/JSON/TSV → Vega charts) | → |
| improvements | createImprovement, searchImprovements, addImprovementEvidence | → |
| plans | createPlan, createPlanPhase, previewTaskDependencyCascade | → |
| knowledge_graph | kg_search, kg_get_entity, kg_related_documents | → |
| organization | createOrganisation, createWorkspace, createProject | → |
| members | inviteMember, updateMemberRole, setMemberActive | → |
| teams | createTeam, grantTeamWorkspaceAccess | → |
| permissions | upsertResourcePermission, setResourcePermissionOverride | → |
| billing | previewSubscriptionChange, applySubscriptionChange, purchaseCreditPackage | → |
| kg_admin | setKgWorkspaceScope, triggerKgRebuild, previewKgRebuild | → |
| settings | getOrgSettings, updateOrgFeatureFlags, updateUserPreferences | → |
Full live catalogue: stablebaseline.io/docs/mcp/tools
Highlights
- End-to-end agent-managed: agents drive sign-up → billing → org/workspace/project lifecycle → members → teams → permissions → Knowledge Graph scope. No human touching the UI.
- Living documents: AI-native rich-text editor (CDMD = Markdown + inline diagrams, data tables, media, version history, real-time multi-author edits).
- Visual diagram creation & editing: 40+ types — AI-generated, diagram-as-code, or freehand.
- Plans, timelines, tasks, improvements & bugs: Day/Week/Month/Quarter/Year zoom Gantt, drag-to-reschedule, dependency cascades.
- Self-learning Knowledge Graph (Enterprise): auto-built from every doc/plan/improvement; read by every connected agent. Stops hallucinations around gaps. Run one brain workspace-wide, one per department, or one per project — RBAC end-to-end.
- Compliance Readiness (Enterprise): scans against OWASP ASVS, NIST SSDF, CIS Controls, ISO 27001, SOC 2, TISAX. Readiness only — not legal advice, not an attestation.
- Bring-your-own-agent — no lock-in: any MCP-compatible client.
- Credit-cost preview ritual for every spending operation:
previewX→ confirm token →applyX.
Auth
| Method | When to use |
|---|---|
| OAuth 2.1 + Dynamic Client Registration | Default for desktop apps and IDE extensions. Smithery handles it automatically. |
Bearer API key (sta_...) | CI/CD, headless agents, server-to-server. Mint at app.stablebaseline.io/settings/mcp-keys. Project-scoped or global. |
OAuth endpoints:
authorize:https://app.stablebaseline.io/oauth/authorizetoken:https://api.stablebaseline.io/oauth/tokenregister(DCR):https://api.stablebaseline.io/oauth/register- Scopes:
org_admin,org_billing,org_members,org_teams,org_permissions,org_settings,kg_admin,lifecycle
Discovery
| Surface | URL |
|---|---|
.well-known/mcp.json (manifest) | https://stablebaseline.io/.well-known/mcp.json |
| Official MCP Registry | https://registry.modelcontextprotocol.io/v0/servers?search=io.stablebaseline |
| Smithery | https://smithery.ai/servers/stablebaseline/sb-mcp |
llms.txt | https://stablebaseline.io/llms.txt |
robots.txt | https://stablebaseline.io/robots.txt |
sitemap.xml | https://stablebaseline.io/sitemap.xml |
security.txt | https://stablebaseline.io/.well-known/security.txt |
Pricing
- Free: 100 one-off credits, no credit card.
- Pro: A$79 / user / month, 500 credits / user / month.
- Enterprise: custom — Knowledge Graph, Compliance Readiness, audit logs, SSO. Contact.
Status
- 🟢 MCP Registry:
io.stablebaseline/sb—active,isLatest. Listed. - 🟢 Smithery: published with
configSchema(Quick Setup). 184 tools / 8 prompts / 6 resources discovered. - 🟢 `.well-known/mcp.json`: live, schema 2024-11-05.
- 🟢 Endpoint:
api.stablebaseline.io/functions/v1/cloud-serve/mcp— accepts POST withmcp-protocol-version: 2025-03-26.
Issues & support
- Bug or feature request: open an issue here.
- Security report: see
.well-known/security.txt— RFC 9116 contact. - General support: https://stablebaseline.io/contact
License
This repository (README, llms-install.md, brand assets, and any documentation) is licensed under the MIT License — see LICENSE. You can copy, fork, and adapt these docs freely.
The Stable Baseline product itself — the SaaS application, the MCP server implementation, and the company brain Knowledge Graph — is proprietary, closed-source, and provided as a service at https://stablebaseline.io under the Stable Baseline Terms of Service.
Built by Orixian Solutions Pty Ltd, Sydney, Australia.