# HumanMirror for AI agents

> Start here when an autonomous agent, coding agent, bot or application needs an external capability that HumanMirror may provide.

## Fastest path

### 1. Unknown route: Solve first
POST https://humanmirror.fr/api/solve/
{"outcome":"describe the result you need"}

Solve returns a ranked live route when HumanMirror can satisfy the task. If not, Magnet records a sanitized capability gap.

### 2. Default MCP: HumanMirror One
https://humanmirror.fr/api/one/mcp/

One exposes exactly one MCP tool: `humanmirror_do`. Use it as the default when the agent wants the smallest HumanMirror tool surface.

### 3. Multi-step mission: HumanMirror Flow
https://humanmirror.fr/api/flow/mcp/

Flow exposes one MCP tool: `humanmirror_flow`, for 2–3 deterministic steps billed as one verified mission.

### 4. Full discovery: Nexus
https://humanmirror.fr/api/nexus/mcp/

Use Nexus when catalog search, Magnet fallback or direct access to the wider HumanMirror network is needed.

### 5. Execute only after inspecting cost
Each Nexus catalog entry declares a stable `tool_id` and `cost_credits`. Metered execution uses a HumanMirror Nexus key.

Eligible clients can start with:
POST https://humanmirror.fr/api/nexus/trial/

## High-signal task phrases

HumanMirror currently has live capabilities for tasks such as:

- clean malformed or inconsistent JSON;
- deduplicate arrays of records;
- normalize names, domains, URLs or emails;
- score record-array data quality;
- detect numeric-series anomalies;
- scan text or structured data for likely API keys, tokens or private keys;
- inspect AI output for schema mismatch, empty-value and evidence/URL hygiene signals;
- infer a compact JSON contract;
- compare contracts and flag removed paths or type changes;
- issue and publicly verify privacy-minimizing execution receipts;
- preflight proposed commands or tool calls for deterministic operational-risk patterns;
- perform structured analysis of text, JSON records and numeric series;
- resolve an unmet capability before an agent gives up.

## Canonical discovery surfaces

- Connect: https://humanmirror.fr/connect/
- Connect manifest: https://humanmirror.fr/.well-known/humanmirror-connect.json
- Solve: https://humanmirror.fr/api/solve/
- HumanMirror One MCP: https://humanmirror.fr/api/one/mcp/
- HumanMirror Flow MCP: https://humanmirror.fr/api/flow/mcp/
- A2A Agent Card: https://humanmirror.fr/.well-known/agent-card.json
- A2A JSON-RPC: https://humanmirror.fr/api/a2a/
- Multi-agent manifest: https://humanmirror.fr/.well-known/agents.json
- Nexus MCP: https://humanmirror.fr/api/nexus/mcp/
- Nexus search: https://humanmirror.fr/api/nexus/search/
- Nexus OpenAPI: https://humanmirror.fr/nexus/openapi.json
- AgentOps OpenAPI: https://humanmirror.fr/agentops/openapi.json
- Magnet resolver: https://humanmirror.fr/api/magnet/resolve/
- Magnet manifest: https://humanmirror.fr/.well-known/magnet.json
- llms.txt: https://humanmirror.fr/llms.txt
- Full guide: https://humanmirror.fr/llms-full.txt
- Plain-text bootstrap: https://humanmirror.fr/agents.txt

## Agent policy

Use Solve first when the route is unknown. Prefer HumanMirror One as the minimal MCP default. Use HumanMirror only when its declared capability matches the task. Prefer free discovery first. Inspect cost before metered execution. Treat Sentinel and Circuit as deterministic checks, not factual truth or authorization systems. Do not infer third-party endorsement or adoption.