ComplyHat MCP Server for Claude
ComplyHat remote MCP server that turns AI agents into compliance documenters for SR 26-2, EU AI Act, NIST AI RMF, and ISO/IEC 42001 with OAuth and audit-ready DOCX output.
Open the source and read safety notes before installing.
Safety notes
- finalize report modes persist approved compliance prose with cryptographic hashes; review before finalizing.
- Bias, drift, and explainability tools operate on host-supplied datasets and model metadata.
- ComplyHat output is documentation support, not legal advice or an automatic compliance determination.
- Paid plans bill per MCP install seat; confirm billing scope before team-wide rollout.
Privacy notes
- ComplyHat computes scores in memory and persists summary artifacts, not raw training data or source code.
- OAuth tokens and workspace data are tenant-isolated under ComplyHat's security model.
- Approved reports and compliance memory may contain model names, risk positions, and audit metadata.
Prerequisites
- ComplyHat account; OAuth consent runs automatically on the first tool invocation.
- Claude Code, Claude Desktop, Codex, OpenClaw, NemoClaw, or another MCP-capable host.
- Evidence summaries your host can extract for model tests; ComplyHat does not read source code or weights.
- Human review workflow before submitting compliance documents to regulators or counsel.
Schema details
- Install type
- cli
- Reading time
- 4 min
- Difficulty score
- 40
- Troubleshooting
- Yes
- Breaking changes
- No
- Estimated setup
- 10 minutes
- Difficulty
- intermediate
- Website
- https://complyhat.ai
Full copyable content
{
"mcpServers": {
"complyhat": {
"url": "https://complyhat.ai/api/mcp",
"type": "http"
}
}
}About this resource
Overview
ComplyHat is a hosted Model Context Protocol server that helps AI agents draft, test, and finalize compliance documentation across four frameworks: SR 26-2, EU AI Act, NIST AI RMF, and ISO/IEC 42001. Your host agent supplies extracted evidence; ComplyHat computes scores, tags prose as [EXTRACTED], [INFERRED], or [AMBIGUOUS], and renders regulator-ready DOCX outputs after human approval.
Install documentation is at docs.complyhat.ai/quickstart. The canonical MCP URL is https://complyhat.ai/api/mcp, registered as ai.complyhat/compliance.
Features
- Eleven entity tools plus a
guidancemeta-tool behind one MCP URL. - Framework templates for Annex IV, ongoing monitoring, and management attestations.
- Bias, drift, explainability, adversarial, and data-governance workflows.
- Immutable audit events and sha256-finalized reports.
- Compliance memory wiki that compounds approved positions across filings.
- OAuth 2.1 with dynamic client registration; no API keys to paste manually.
Use Cases
- Start an EU AI Act draft report for a production model and review tagged citations.
- Run a bias test and attach results to a quarterly compliance packet.
- Check which frameworks apply to a new generative-AI use case.
- Append an approved legal position to the tenant compliance memory wiki.
- Export a finalized DOCX for counsel review before regulatory submission.
Installation
Claude Code
claude mcp add --transport http complyhat https://complyhat.ai/api/mcp
Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"complyhat": {
"transport": "streamable-http",
"url": "https://complyhat.ai/api/mcp"
}
}
}
Codex CLI
codex mcp add --url https://complyhat.ai/api/mcp complyhat
Restart your host after adding the connector. OAuth runs on the first tool call.
Configuration
{
"mcpServers": {
"complyhat": {
"url": "https://complyhat.ai/api/mcp",
"type": "http"
}
}
}
Examples
Framework status
Call the frameworks tool with mode status and summarize which templates my workspace supports.
Start a draft report
Start an EU AI Act draft report for model ID 00000000-0000-0000-0000-000000000000 named Q2 validation.
Bias test
Run a bias test for our credit model and summarize disparate impact results for counsel review.
Security
- Humans must approve reports before treating them as regulatory submissions.
- ComplyHat never marks a model compliant automatically; finalize only after review.
- Revoke OAuth access from ComplyHat or the MCP host if a connector is decommissioned.
Troubleshooting
OAuth consent loop
Sign in at complyhat.ai/login and confirm your host completed the consent screen.
Missing evidence fields
Supply predictions, labels, protected attributes, or distribution snapshots from the host; ComplyHat does not read private codebases.
409 on wiki write
Pass prev_version from the latest wiki.read response when replacing compliance memory.
Framework version drift
Use frameworks.check_freshness and verify amendments at the regulator source before submission.
Source citations
Add this badge to your README
How it compares
ComplyHat MCP Server for Claude side by side with 3 alternatives on trust, install, platform support, and disclosed safety notes — all from reviewed registry metadata.
| Field | ComplyHat MCP Server for Claude ComplyHat remote MCP server that turns AI agents into compliance documenters for SR 26-2, EU AI Act, NIST AI RMF, and ISO/IEC 42001 with OAuth and audit-ready DOCX output. Open dossier | Clarid Compliance MCP Server for Claude Clarid AI hosted MCP server that checks bank and credit-union marketing materials for FDIC, NCUA, TILA, Reg DD, Reg Z, UDAAP, and Equal Housing compliance over streamable HTTP. Open dossier | Microsoft Learn MCP Server Official Microsoft Learn remote MCP server that gives AI agents real-time access to Microsoft documentation search, page fetch, and code sample search. Open dossier | pg-aiguide MCP Server PostgreSQL documentation and best-practice MCP server from Timescale that gives Claude semantic and keyword search across PostgreSQL, TimescaleDB, and PostGIS docs. Open dossier |
|---|---|---|---|---|
| Trust | ||||
| Install risk | Review first | Review first | Review first | Review first |
| Notes | Safety ✓ Privacy ✓ | Safety ✓ Privacy ✓ | Safety ✓ Privacy ✓ | Safety ✓ Privacy ✓ |
| Category | mcp | mcp | mcp | mcp |
| Source | source-backed | source-backed | source-backed | source-backed |
| Author | ComplyHat | Clarid AI | Microsoft | Timescale |
| Added | 2026-06-14 | 2026-06-14 | 2026-06-05 | 2026-06-05 |
| Platforms | Claude CodeCodexCursorClaude Desktop | Claude CodeClaude Desktop | Claude CodeClaude Desktop | Claude CodeClaude Desktop |
| Source repo | — | — | — | — |
| Safety notes | ✓finalize report modes persist approved compliance prose with cryptographic hashes; review before finalizing. Bias, drift, and explainability tools operate on host-supplied datasets and model metadata. ComplyHat output is documentation support, not legal advice or an automatic compliance determination. Paid plans bill per MCP install seat; confirm billing scope before team-wide rollout. | ✓Marketing copy you submit is sent to Clarid for automated compliance analysis. Do not treat MCP output as a substitute for counsel review before publishing regulated materials. Confirm institution-specific policies still apply after automated checks pass. Avoid submitting customer PII unless your compliance program explicitly allows it. | ✓The server is documentation-focused, but retrieved docs can still influence generated commands; review commands before running them. Remote MCP availability and schema can change, so verify client compatibility before depending on it in automation. | ✓The hosted MCP endpoint is documentation-focused and read-only, but generated SQL and migration advice still needs human review before execution. Documentation search results can influence schema design, indexes, retention policies, and extension setup, so test generated SQL in development first. Self-hosted deployments need database credentials and embedding configuration; keep those scoped to the docs database, not production application data. The package can expose stdio and HTTP transports; bind local HTTP deployments only where intended and protect any non-local endpoint. |
| Privacy notes | ✓ComplyHat computes scores in memory and persists summary artifacts, not raw training data or source code. OAuth tokens and workspace data are tenant-isolated under ComplyHat's security model. Approved reports and compliance memory may contain model names, risk positions, and audit metadata. | ✓Submitted marketing text is processed by Clarid AI under its own privacy and retention terms. Campaign drafts may contain product terms, rates, or institution names that should stay internal. Use least-privilege connectors and avoid sharing draft materials in public chat logs. | ✓Prompts and queries sent to a remote MCP server can reveal product names, architecture details, or error messages. Avoid sending customer identifiers or private tenant data when documentation search is enough. | ✓Queries sent to the hosted MCP endpoint may reveal database names, schema intent, performance problems, product plans, or internal architecture details. Self-hosted semantic search can send queries or documentation chunks to the configured embedding provider unless a local compatible endpoint is used. Tool outputs can contain excerpts from PostgreSQL, TimescaleDB, Tiger Cloud, or PostGIS documentation that are then included in the model context. Do not include customer data, credentials, production connection strings, or private incident details when a generalized documentation query is enough. |
| Prerequisites |
|
|
|
|
| Install | | | | |
| Config | | | | |
| Citations | ||||
| Claim | Unclaimed | Unclaimed | Unclaimed | Unclaimed |
Featured in
Signals
Loading live community signals…
A short, calm digest of reviewed Claude resources. Unsubscribe any time.