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

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 apiendpoint, apimethod, description, samplerequest, sampleresponse, expected_behavior.

Security note: contentfromuntrusted_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 runapiaudit. 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>" } }

getauditresults returns per-model accuracy scores and hallucination examples. getremediationsuggestions returns targetfile, patchstrategy, 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/langchaintools.py](scripts/langchaintools.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](references/REFERENCE.md) for copy-paste config blocks for Claude Desktop, VS Code Copilot, Cursor, and Continue.dev.

Error Reference