FLOCORE Governed MCP
Governed MCP front door to FLOCORE; write calls need human approval and signed receipts.
Should I use this
Quality & Safety
Findings (2)
- HIGHTool poisoning patterns detected
- MEDIUMTool description contains URL to non-standard domainin flocore_support_ask
Based on automated analysis of tool definitions and protocol compliance.
Context Cost
This is the approximate number of tokens consumed each time the server's tools are loaded into a model's context. Higher counts reduce the attention available for other tasks.
Install
One-Click Install
Add this to your `claude_desktop_config.json` file:
{
"mcpServers": {
"flocore": {
"url": "https://fo.flocore.tech/mcp"
}
}
}Remote endpoints
https://fo.flocore.tech/mcpstreamable-httpWhat it can do
Tool inventory
Tools (4)
🟢aegis_inspect(text, agent, referred_by)
Screen a prompt or text for injection, jailbreak and exfil intent. Free, public, defensive-only: returns a risk score (0..1), the detected intents, and whether it is suspicious. No token needed, no tenant data. A pattern screen: it misses paraphrases, so a clean result is not a guarantee. An anonymous call stores nothing; if you pass an agent name, that name and the score are recorded.
Input Schema
{
"type": "object",
"properties": {
"text": {
"type": "string",
"description": "the prompt or text to screen"
},
"agent": {
"type": "string",
"description": "optional: your agent name or a contact; if given, it is recorded with the score"
},
"referred_by": {
"type": "string",
"description": "optional: the agent id that referred you, so the referral is attributed to them"
}
},
"required": [
"text"
],
"additionalProperties": false
}🟡flocore_onboard_request(agent, owner_contact, rehearsal_only, org, intended_use, ...)
Registers an onboarding request for human review (human-gated approval, Ed25519 signed receipts, EU AI Act Article 12 audit evidence). Public: no token needed. Returns a request id and a sandbox token (fc_test_...) immediately, to rehearse the loop; a live token is issued only after a person approves the application. Set rehearsal_only to true to get a sandbox token without applying.
Input Schema
{
"type": "object",
"properties": {
"agent": {
"type": "string",
"description": "your agent name or identifier"
},
"owner_contact": {
"type": "string",
"description": "the email address of the person who owns the agent, so a person can approve it and be sent its token (required unless rehearsal_only is true)"
},
"rehearsal_only": {
"type": "boolean",
"description": "optional: true to get ONLY a sandbox token to rehearse the approval loop. Nothing is registered, no email is sent and no one is asked to review anything. Limits: 3 per hour per connection. To apply for real, call again without it and give owner_contact"
},
"org": {
"type": "string",
"description": "optional: the organisation the agent acts for"
},
"intended_use": {
"type": "string",
"description": "optional: what world-affecting actions you want governed"
},
"referred_by": {
"type": "string",
"description": "optional: the governed agent or partner that sent you here (recorded as the referral)"
},
"jurisdiction": {
"type": "string",
"description": "optional: the legal jurisdiction that brought you here, a code from the Advatar pitch (eu, coe, za, kr, us, vn, ng, ke, mx, oecd). Recorded so we learn which law matters to builders"
},
"source": {
"type": "string",
"description": "optional: where you found us, for example a registry or the pitch (letters, digits, dots and dashes)"
},
"public_key": {
"type": "string",
"description": "optional: your agent's own Ed25519 PUBLIC key, raw 32 bytes as base64url or a public JWK as JSON text. Give it and, once a human approves you, your token is bound to it and your key's thumbprint appears on your signed receipts. Never send a private key."
}
},
"required": [
"agent"
],
"additionalProperties": false
}🟢flocore_support_ask(question)
Ask a plain-English support/product question about FLOCORE itself, agent to agent, instead of a helpdesk. Public, no token needed. Answered only from FLOCORE's own real, published, reproducible claims: an answer not supported by them is refused, never invented. Privacy: your question (max 500 characters) is sent to a hosted language model provider, currently Groq in the United States, to choose which published sentences answer it; do not include personal information. See https://humanseal.world/privacy.html.
Input Schema
{
"type": "object",
"properties": {
"question": {
"type": "string",
"description": "your plain-English question about FLOCORE"
}
},
"required": [
"question"
],
"additionalProperties": false
}🟢flocore_verify_counterparty(agent_slug, receipt_canonical, receipt_signature_ed25519)
Before transacting with another agent, check whether it is FLOCORE-governed (a human-approval gate on its world-affecting actions, Ed25519 signed receipts) and, if it handed you a receipt, verify that receipt independently. Public, no token needed. A sandbox receipt is reported as a rehearsal and never as valid proof.
Input Schema
{
"type": "object",
"properties": {
"agent_slug": {
"type": "string",
"description": "the counterparty's FLOCORE agent/tenant slug"
},
"receipt_canonical": {
"type": "string",
"description": "optional: the receipt's `canonical` field, if the counterparty handed you one to check"
},
"receipt_signature_ed25519": {
"type": "string",
"description": "optional: the receipt's `signature_ed25519` field, required together with receipt_canonical to verify it"
}
},
"required": [
"agent_slug"
],
"additionalProperties": false
}Community
Evidence