creedspace-mcp
OfficialExposes Creed Space safety tools as MCP tools in LangChain, allowing agents to perform constitutional adjudication and response attestation.
Integrates with OpenAI Agents SDK (via HTTP transport) and OpenAI Codex (via STDIO transport) to validate responses and reject harmful prompts based on selected personas.
creedspace-mcp-server
Universal MCP server for Creed Space - AI safety guardrails in 10 seconds.
Available in
The server is published to the canonical MCP catalogues, so registry-aware clients can discover and install it directly — or use any command in Quick Start below.
npm —
@creedspace/mcp-server, the package every install path resolves to.Official MCP Registry —
io.github.Creed-Space/creedspace-mcp-server.GitHub — source, issues, and the
server.jsonmanifest.Smithery —
nell/creedspace, a gateway-fronted Streamable HTTP endpoint.
Also rolling out across the wider MCP ecosystem: mcp.directory, mcpservers.org, PulseMCP (via the registry ingest), and mcp.so.
Related MCP server: repo-seatbelt
Quick Start
# STDIO transport (Claude Desktop, OpenAI Codex)
npx @creedspace/mcp-server --persona ambassador
# HTTP transport (OpenAI Agents SDK)
npx @creedspace/mcp-server --transport http --port 3100
# Test API connection
npx @creedspace/mcp-server testWhat is Creed Space?
Creed Space provides personalized AI safety guardrails through Constitutional AI personas. Each persona enforces specific values and behaviors, ensuring AI assistants operate within defined ethical boundaries.
🛡️ Reduces harmful AI outputs via constitutional evaluation of every response
🎯 Refuses dangerous prompts according to the active persona's values
🚀 10-second setup with any MCP-compatible AI
Available Personas
Persona | Icon | Purpose |
Ambassador | 🤝 | Professional communication |
Nanny | 👶 | Child-safe interactions |
Sentinel | 🛡️ | Privacy and security focus |
Godparent | 🕊️ | Religious and ethical guidance |
Muse | 🎨 | Creative exploration |
Anchor | ⚓ | Reality grounding |
Installation
Option 1: Use with npx (Recommended)
# No installation needed - just run!
npx @creedspace/mcp-server --persona ambassadorOption 2: Global Installation
npm install -g @creedspace/mcp-server
creedspace-mcp --persona ambassadorOption 3: Project Dependency
npm install @creedspace/mcp-serverClaude Desktop Integration
Add to your claude_desktop_config.json:
{
"mcpServers": {
"creedspace": {
"command": "npx",
"args": ["@creedspace/mcp-server", "--persona", "ambassador"]
}
}
}Then restart Claude Desktop to load the Creed Space guardrails.
Configuration
Environment Variables
# .env file
CREEDSPACE_API_URL=https://api.creed.space
CREEDSPACE_API_KEY=your-api-key-here # Optional
CREEDSPACE_DEFAULT_PERSONA=ambassadorCommand Line Options
creedspace-mcp \
--persona ambassador \
--url https://api.creed.space \
--api-key YOUR_KEY \
--cache-ttl 300000 \
--offline
# HTTP Transport Options
creedspace-mcp \
--transport http \
--port 3100 \
--host localhost \
--api-key YOUR_KEY \
--cors \
--cors-origin http://localhost:3000 \
--statelessTransport Types
Transport | Use Case | Platforms |
| Local subprocess communication | Claude Desktop, OpenAI Codex |
| HTTP server for remote/local connections | OpenAI Agents SDK, custom integrations |
Configuration File
# Generate example configs
creedspace-mcp --generate-config
# Use config file
creedspace-mcp --config creedspace.jsonAvailable MCP Tools
The server provides 16 tools to MCP clients:
Constitutions
get_constitution- Get the merged constitution for a personaget_constitution_by_id- Get a specific constitution by IDsearch_constitutions- Search the constitution library
Personas
list_personas- List all available personasset_persona- Switch the active personaget_active_persona- Get the currently active personaget_uvc_qualities- Get desired/disliked/never qualitiesget_system_prompt- Get a complete persona system promptpreview_export- Preview the export configuration
Runtime guardrails
adjudicate- Get a policy decision kernel for a requestattest_response- Validate a response against the active creedget_anchor- Get a compact non-negotiable-rules anchorheartbeat- Periodic re-anchoring to prevent context driftclear_cache- Clear the local cache
Multi-scale value handshake
perform_multi_scale_handshake- N-party value handshake across micro/meso/macro scalesget_scale_attestation- Get an attestation record with hash chain
Programmatic Usage
import { CreedSpaceMCPServer } from '@creedspace/mcp-server';
// Start server programmatically
const server = new CreedSpaceMCPServer({
persona: 'ambassador',
apiUrl: 'https://api.creed.space',
cacheEnabled: true
});
await server.start();// Use the API client directly
import { CreedSpaceClient } from '@creedspace/mcp-server';
const client = new CreedSpaceClient();
const personas = await client.getPersonas();
const constitution = await client.getMergedConstitution('ambassador');Platform Integration Examples
OpenAI Agents SDK (HTTP Transport)
from agents import Agent
from agents.mcp import MCPServerStreamableHttp
# Start the server first:
# npx @creedspace/mcp-server --transport http --port 3100
server = MCPServerStreamableHttp(
url="http://localhost:3100/mcp",
name="creedspace"
)
agent = Agent(
name="safe-agent",
tools=[server.get_tools()]
)
# The agent now has access to Creed Space safety toolsOpenAI Codex (STDIO Transport)
Add to ~/.codex/config.toml:
[mcp_servers.creedspace]
command = "npx"
args = ["@creedspace/mcp-server", "--persona", "ambassador"]
[mcp_servers.creedspace.env]
CREEDSPACE_API_URL = "https://api.creed.space"VS Code / Cursor
{
"mcp.servers": {
"creedspace": {
"command": "npx",
"args": ["@creedspace/mcp-server", "--persona", "ambassador"]
}
}
}Continue.dev
{
"models": [{
"provider": "openai",
"mcp_servers": [{
"command": "npx",
"args": ["@creedspace/mcp-server"]
}]
}]
}LangChain
from langchain.tools import MCPTool
creedspace = MCPTool(
command="npx",
args=["@creedspace/mcp-server", "--persona", "ambassador"]
)Testing
# Test API connection
npx @creedspace/mcp-server test
# Test with specific URL
npx @creedspace/mcp-server test --url http://localhost:8000Offline Mode
The server includes intelligent caching for offline usage:
# Enable offline mode with cached data
creedspace-mcp --offline --persona ambassadorDevelopment
# Clone the repository
git clone https://github.com/Creed-Space/creedspace-mcp-server.git
cd creedspace-mcp-server
# Install dependencies
npm install
# Build TypeScript
npm run build
# Run in development mode
npm run dev
# Run tests
npm testAPI Documentation
Full API documentation available at https://api.creed.space/api-docs
Support
🌐 Website: https://www.creed.space
📧 Email: support@creed.space
🐛 Issues: GitHub Issues
💬 Discord: Join our community
License
MIT © Nell Watson
Building critical AI safety infrastructure that shapes autonomous AI-human value interaction.
Available Tools
16 toolsadjudicateC
Get a policy decision kernel for a user request based on active constitutions and persona
Inputs: question (required), context, persona_id. Returns: Policy decision kernel returned by the Creed Space PDP
attest_responseARead-only
Validate a response against the active creed to check for violations and ensure attestation
Inputs: response (required), persona_id. Returns: status, warnings, creedHash, violations, +1 more
clear_cacheAIdempotent
Clear the local cache to force fresh data from the API
Inputs: no named parameters. Returns: cleared, message
get_active_personaARead-only
Get the currently active persona for this session
Inputs: no named parameters. Returns: id, icon, name, description
get_anchorARead-only
Get a compact anchor (1-2KB) with top 10 non-negotiable rules and creed hash for quick context reinforcement
Inputs: max_length, persona_id. Returns: anchor, persona, maxLength, personaId, +2 more
get_constitutionARead-only
Get the merged constitution for a specific persona, including all active rules and UVC qualities
Inputs: persona_id. Returns: icon, content, persona, uvcToken, +2 more
get_constitution_by_idARead-only
Get a specific constitution by its ID
Inputs: constitution_id (required). Returns: id, name, content, personaId, +2 more
get_scale_attestationARead-only
Get attestation record with hash chain for a specific scale and entity
Inputs: scale (required), entity_id (required), include_chain. Returns: Attestation record, attestation chain, or null-attestation placeholder
get_system_promptARead-only
Get a complete system prompt for a persona including constitution and UVC qualities
Inputs: persona_id. Returns: personaId, systemPrompt, creedAttestation
get_uvc_qualitiesARead-only
Get the desired, disliked, and never qualities for a persona from the UVC system
Inputs: persona_id. Returns: personaId, qualities, configured
heartbeatBRead-only
Periodic re-anchoring to prevent context drift - returns mini-anchor every 10 messages
Inputs: force, persona_id, message_count. Returns: text, persona, anchored, personaId, +1 more
list_personasARead-only
List all available AI personas with their descriptions and current status
Inputs: no named parameters. Returns: current, available
perform_multi_scale_handshakeB
Perform N-party value handshake across micro/meso/macro scales with anti-collusion and hierarchical budgets
Inputs: parties (required), invariants. Returns: Handshake result returned by the Creed Space safety stack
preview_exportC
Preview the export configuration for Claude Code or other AI tools
Inputs: persona_id, include_uvc, include_constitutions, include_system_prompt. Returns: preview, personaId, includeUvc, personaName, +3 more
search_constitutionsARead-only
Search constitutions by keyword or persona
Inputs: query, persona_id. Returns: query, persona, results, constitutions
set_personaAIdempotent
Switch to a different AI persona for the current session
Inputs: persona_id (required). Returns: id, icon, name, message, +1 more
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 16 tools
Tools have distinct purposes, but several 'get' variants (constitution, qualities, system prompt, anchor) overlap conceptually. Descriptions clarify differences, so confusion is limited.
Naming mixes verb_noun patterns (get_constitution, list_personas) with standalone verbs (adjudicate, heartbeat) and verbose compounds (perform_multi_scale_handshake). This inconsistency reduces predictability.
At 16 tools, the count is slightly above the typical well-scoped range of 3-15. The domain is broad, but a few tools (e.g., get_constitution vs. get_constitution_by_id) could be consolidated.
The set covers core operations for personas, constitutions, and attestations, including retrieval, search, active persona selection, and validation. Missing create/update/delete may be acceptable depending on the intended use case.
Maintenance
Related MCP Connectors
AgentGuard — 20-tool AI safety MCP: policy preflight, risk scoring, audit logging, rate limits.
Real-time MCP & A2A prompt-injection, jailbreak, and tool-schema poisoning firewall for AI agents.
Real-time prompt injection filtering, MCP payload inspection, and A2A trust scoring.
Jailbreak-proof AI guardrails. Automated Reasoning SMT solver, not an LLM. ZK proofs included.
Related MCP Servers
- AlicenseAqualityDmaintenanceUnified MCP safety server that detects prompt injection (75 patterns), scans LLM outputs for leaked secrets/PII, enforces API cost budgets, and creates signed audit trails. Zero ML dependencies, pure Python.1765 PyPI1MIT
- AlicenseNot gradedqualityDmaintenanceRuntime safety guardrails for AI coding agents. Checks file access, validates shell commands, and scores your repo's AI safety — all via MCP.9 npm8MIT
- AlicenseNot gradedqualityDmaintenanceMCP server for AI agent security guardrails. Provides input validation, prompt injection detection, PII redaction, output filtering, policy enforcement, rate limiting, and comprehensive audit logging.72 npm1MIT
- AlicenseNot gradedqualityFmaintenanceConstitutional MCP server enforcing 13 Floors of governance for AI agents, providing tools for session anchoring, reasoning, safety critique, and audit logging.AGPL 3.0