Use the text tools from your AI client

Use the Watermark-free AI Humanizer, watermark remover, watermark comparison and AI Detector through the same account. Rewrites use your account's monthly allowance: $5 a month gives 100,000 characters. Detector checks are free and limited to 100 per account each UTC day.

I tested the single-pass rewrite method with the published SynthID Text detector; the prompt and the sampling temperature of the current tools were not part of that test. The method is the strongest available removal approach: replacing the wording throughout. Read the results and examples before deciding whether it fits your writing.

Best effort: the ChatGPT, Claude and Gemini watermark detectors are not public, so complete removal cannot be verified.

Connect by URL

Add https://painintheagent.com/mcp as a remote MCP server in your client. Follow its sign-in prompt, open your existing website account, and review the permissions. The connection uses Authorization Code with S256 PKCE.

Clients that accept a bearer header can also use an account API key with this URL. Agents using a claim code can follow auth.md. A claim must be confirmed by the account owner before the agent can read saved rewrites or start work.

Remote requests run in the background. Generate a UUID run_id before starting a tool, keep it, and use get_result to retrieve progress or the result. Reuse that ID for the same operation and input after a lost response. The earlier get_rewrite name still works.

Choose a remote MCP tool

• start_rewrite: remove the watermark with one rewrite. Paid runs include one meaning check by a second model. Possible changes are reported in review without editing the rewrite. Check review_status: unavailable means the rewrite is ready but its meaning check could not finish. Free runs and older unchecked results carry review_method: not_run.
• start_humanize: humanize text and remove its SynthID-style watermark. Watermark-free rewriting is included; watermark_removal remains optional for compatibility.
• start_watermark_comparison: compare source_text and candidate_text for meaning changes. Private watermark status remains unverifiable.
• start_ai_detection: highlight AI-writing patterns in at least 50 words of prose, with the share of words they cover. Short excerpts are less reliable.
• get_limits and get_result: read the rewrite allowance or an existing result without using another rewrite.

Install the local MCP package

The published local package currently exposes the rewrite workflow. Use the remote URL above for all four text tools.

1. Create a key above after signing in with your email.
2. Download the latest release. For Claude Desktop, open the MCPB bundle and enter your key in the extension settings.
3. For Codex, Claude Code, Cursor or another client that supports local MCP servers, follow the client configuration instructions. The bundle includes its dependencies and requires Node.js 22 or later.

The repository also contains plugins for Codex and Claude Code. Installation from the repository is separate from approval in a client's public directory.

The local package is published in the official MCP Registry as io.github.krllagent/painintheagent-mcp.

Ask for a rewrite

For example, ask your agent to remove the watermark from a draft through Pain in the Agent and review any reported changes to claims, conditions and numbers: paid start_rewrite runs include one meaning check. The agent calls start_rewrite, saves the run ID, and uses get_result to check progress. A rewrite and its meaning check can take several seconds or longer, depending on length and provider availability; start_humanize writes several drafts and can take a few minutes.

For the local package, keep the client open while it works. With a remote connection, the queued job continues after the client disconnects. If a connection drops, use the same run ID to check for a saved result. Starting a new request starts new work. The server rejects reuse of an ID with different input.

Direct API access

Send your key as an Authorization: Bearer header to https://painintheagent.com. The API reference includes the request format and errors. The OpenAPI description provides the same operations in a machine-readable format.

• GET /api/v1/watermark/limits reads the account's rewrite allowance.
• POST /api/v1/watermark/run accepts operation: watermark (default), humanize, compare or detect-ai, and streams progress as NDJSON.
• GET /api/v1/watermark/run?id=UUID retrieves progress or a saved result.

All four tools use this endpoint and the same account key. A humanizer request always includes watermark-free rewriting; the optional watermark_removal field remains accepted for compatibility. A comparison requires candidate_text alongside the original text. These options and the operation are part of the run ID's identity.

The client supports PAININTHEAGENT_BASE_URL for a local API gateway. It uses the configured endpoint for every request and does not fall back to another provider.

Resources for agents

Public pages also respond to Accept: text/markdown. These resources describe the API and the installed MCP workflow.

• API catalog
• Agent skills index and rewrite instructions
• AI catalog
• Remote MCP Server Card and authorization discovery
• Web Bot Auth public key directory

Your text and keys

The service processes and stores submitted text under the same privacy policy as the web tool. Only a hash of each API key is stored on the server. A key can start rewrites and read your account's saved runs; it cannot manage keys. You can revoke it above.