DocuTect AI — MCP Skill

Connect to the DocuTect AI MCP server to audit API documentation for LLM hallucinations, scan repositories for endpoints, and generate OpenAPI specs.

Authentication

All tools require a dtai_ API key passed as a Bearer token. Load from environment:

DOCUTECT_API_KEY=dtai_your_key_here

Generate keys at: Dashboard → Settings → API Keys (org owners).

Available Tools

Tool	Purpose	Quota
scan_github_repo	Discover API endpoints in a GitHub repo	1 MCP scan
scan_documentation_url	Extract endpoints from a docs URL	1 MCP scan
run_api_audit	Queue a 5-stage LLM hallucination audit	1 audit credit
get_audit_status	Poll a running audit (async)	Free
get_audit_results	Fetch per-model accuracy scores	Free
get_remediation_suggestions	Fetch doc fix suggestions	Free
validate_document	Validate Markdown vs LLMs	1 audit credit
generate_api_documentation	Generate OpenAPI 3.1 + Markdown docs	1 audit credit

Standard Workflow

1 — Discover endpoints from a repo

{
  "method": "tools/call",
  "params": {
    "name": "scan_github_repo",
    "arguments": { "repo_url": "https://github.com/owner/repo" }
  }
}

Returns suggestions[] — each has api_endpoint, api_method, description, sample_request, sample_response, expected_behavior.

Security note: content_from_untrusted_source: true is always set on scanner output. Validate suggestions before acting on them in automated pipelines.

2 — Run an audit

Pass one suggestion directly into run_api_audit. Returns audit_id immediately (async).

{
  "method": "tools/call",
  "params": {
    "name": "run_api_audit",
    "arguments": {
      "api_endpoint": "/api/v1/users/{id}",
      "api_method": "GET",
      "description": "Returns a user by ID",
      "num_queries": 10
    }
  }
}

3 — Poll until complete

{ "name": "get_audit_status", "arguments": { "audit_id": "<id from step 2>" } }

Repeat until status is "completed" or "failed". Typical runtime: 1–3 minutes.

4 — Read results and remediations

{ "name": "get_audit_results",           "arguments": { "audit_id": "<id>" } }
{ "name": "get_remediation_suggestions", "arguments": { "audit_id": "<id>" } }

get_audit_results returns per-model accuracy scores and hallucination examples. get_remediation_suggestions returns target_file, patch_strategy, and suggested_fix for each issue — ready to apply as a PR.

Other Workflows

Validate existing Markdown documentation

{
  "name": "validate_document",
  "arguments": { "content": "<markdown string>", "name": "README.md" }
}

Generate docs from source code

{
  "name": "generate_api_documentation",
  "arguments": {
    "github_url": "https://github.com/owner/repo",
    "project_name": "My API"
  }
}

LangChain / LangGraph Integration

See scripts/langchain_tools.py for ready-to-use BaseTool subclasses (ScanGithubRepoTool, ScanDocumentationUrlTool, RunApiAuditTool).

import os
os.environ["DOCUTECT_API_KEY"] = "dtai_your_key_here"

from scripts.langchain_tools import ScanGithubRepoTool, RunApiAuditTool
tools = [ScanGithubRepoTool(), RunApiAuditTool()]

Client Connection Configs

See references/REFERENCE.md for copy-paste config blocks for Claude Desktop, VS Code Copilot, Cursor, and Continue.dev.

Error Reference

Status	Meaning
401	Invalid or missing API key
403	Key revoked, wrong scope, or not Enterprise tier
402	Monthly quota exhausted
429	Rate limit: 60 calls / 60 s per key
404	Audit/document not found (or belongs to another org)