Pain in the Agent

io.github.krllagentPublished by GitHub account krllagentv0.3.2Updated Nov 5, 2026

Remove text watermarks, humanize and check AI writing with your Pain in the Agent account.

VerifiedSTDIODesktop onlyMedia & DesignAI & ML

Overview

AI-generated overview

Text tooling for AI clients: rewrite text to remove watermarks, humanize drafts, compare versions and score AI writing, using a paid Pain in the Agent account.

What you can do
  • Start watermark-removing rewrites and humanizing passes on supplied text, with before-and-after AI scores
  • Compare a source and candidate text for meaning changes
  • Score text of at least 80 words with an AI detector and highlight AI-writing patterns
  • Poll progress or fetch saved results for any run by its ID
  • Read the account's character allowance, per-text limit and paid-month end date
When to use it
Useful when you want an assistant to rewrite or humanize drafts, check whether a rewrite kept its meaning, or get an AI-writing score, without leaving the chat client. It suits users who already have or will buy a Pain in the Agent plan for the paid rewrites.
Requirements
Runs as a local stdio process on the user's machine (desktop only), via an MCPB bundle or a Node.js 22+ checkout. Needs a Pain in the Agent account and API key, supplied as PAININTHEAGENT_API_KEY; an optional PAININTHEAGENT_BASE_URL points at a gateway. Rewrites require the monthly plan; network access to the hosted service is needed.
Before you install
Rewrites consume characters from a paid monthly plan ($5 for 100,000 characters, up to 10,000 per text); only successful changed results are charged, and pending requests reserve characters. The API key is a secret and should stay in client secret settings or environment, not in conversations or source control. Submitted text goes to the hosted service and its model provider and is stored per its privacy policy; do not submit secrets or confidential personal data. The client cannot verify…

Installation

In SourceWeft

  1. Open Pain in the Agent in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

Pain in the Agent MCP

Use the Pain in the Agent text tools from an AI client: remove text watermarks, humanize drafts, compare versions or check AI writing. Rewrites use the characters of your account's monthly plan: $5 a month for 100,000 characters, up to 10,000 per text, shared with the website and the remote MCP. Only a successful changed result is charged. Comparisons and AI checks are free and limited to 100 a day; without a plan they take texts of up to 1,000 characters, with it up to 10,000. The website's free tries (3 per browser) do not apply to the API or this client.

The research and examples describe tests of rewrite methods against reference SynthID Text watermarks. Those measurements are separate from the live tool workflow. This client cannot verify Claude or Gemini's private watermark keys. AI writing signals are experimental style observations, not proof of authorship.

Installation

First subscribe and create an API key while signed in; rewrites need the monthly plan on that account. Keep the key in your client's secret settings or environment, outside conversations and source control. You do not need an OpenRouter or OpenAI key.

Claude Desktop

Download the .mcpb file from the latest release, open it in Claude Desktop, and enter the service key in its API key setting. The MCPB bundle contains its Node dependencies. It uses Node.js 22 or later.

Codex

Clone this repository or extract a release. The committed dist/index.cjs is ready to run without npm install. Add this to your Codex MCP configuration, replacing the example absolute path with your checkout:

toml
[mcp_servers.painintheagent]command = "node"args = ["/absolute/path/painintheagent-mcp/dist/index.cjs"]env_vars = ["PAININTHEAGENT_API_KEY", "PAININTHEAGENT_BASE_URL"]startup_timeout_sec = 15tool_timeout_sec = 40

Set PAININTHEAGENT_API_KEY in the environment that starts Codex. Omit PAININTHEAGENT_BASE_URL for the public service. Restart the client after changing its environment.

A Codex plugin is also included at plugins/painintheagent-mcp/, with a repository marketplace at .agents/plugins/marketplace.json. To add this source, run codex plugin marketplace add krllagent/painintheagent-mcp and select Pain in the Agent from that source in the app. The generated catalog is named personal; it is a repository source, separate from OpenAI's public plugin directory. The plugin forwards the same two environment variables.

Claude Code

Set PAININTHEAGENT_API_KEY in the environment that starts Claude Code. Then use:

text
/plugin marketplace add krllagent/painintheagent-mcp/plugin install painintheagent-mcp@painintheagent

The plugin includes a rewrite skill and the MCP server. A local checkout can also be loaded with claude --plugin-dir /absolute/path/painintheagent-mcp/claude-plugins/painintheagent-mcp.

Cursor and other local MCP clients

Configure a stdio server with command node and one argument, the absolute path to dist/index.cjs. Supply PAININTHEAGENT_API_KEY through your client's environment or secret setting. Clients have different environment-variable substitution syntax, so use the format your client documents. The MCPB and source release contain the same server.

Tools

ToolWhat it does
start_rewriteStarts a watermark-removing rewrite and promptly returns a run_id. Uses characters from the plan. slop_removal is accepted and has no effect.
start_humanizeHumanizes the text and removes its watermark by replacing the wording throughout, with before-and-after AI scores. Uses characters from the plan.
start_watermark_comparisonCompares source_text and candidate_text for meaning changes. Free within the daily check limit. Does not detect vendor watermarks.
start_ai_detectionScores at least 80 words with a trained AI detector (0–100) and highlights AI-writing patterns. Free within the daily check limit.
get_resultReads progress or the saved result for any text tool without starting model work.
get_rewriteReturns progress, a saved result with fidelity notes, or an error for that ID. It starts no model work.
get_limitsReads the account's character allowance, the per-text limit and when the paid month ends.

Ask your agent to use the tool you need and show its source notes or quoted style evidence. Keep the client open while it works; deep rewriting can take several minutes. The local MCP process holds the streaming connection. Closing it can interrupt the run. The separate remote MCP endpoint at https://painintheagent.com/mcp supports OAuth and continues queued jobs after disconnecting.

Save the request ID. If the stream fails, use get_result with that same ID. The earlier get_rewrite name still works. The client never automatically retries a POST. Reusing an ID with the same operation and input retrieves that run; a different operation, candidate or option is rejected. Failed runs use no characters. Pending requests reserve their characters until they finish or reach the server's deadline, and their IDs cannot start fresh work.

Privacy and gateways

Only supplied text is sent. This client does not read your files, save prose to disk, or emit telemetry. The hosted service sends prose to its model provider and stores source text and results as described in the privacy policy. Your MCP host may retain tool inputs and outputs under its own settings. Do not submit secrets or confidential personal data.

The client reads PAININTHEAGENT_API_KEY and optional PAININTHEAGENT_BASE_URL. The default base URL is https://painintheagent.com. A configured gateway receives every request; there is no credential or endpoint fallback. HTTP is accepted only for localhost gateways. Authentication uses a bearer header, and redirects are rejected.

Development

sh
npm ci --ignore-scriptsnpm run buildnpm testnode scripts/package.mjs

The tests use fictional text and a local fake HTTP service. They cover endpoint selection, failed authentication, request reuse, dropped streams, and a fresh bundled installation speaking MCP over stdio. They do not call a paid model or claim to be live-provider validation.

See the API reference and distribution status. The client is MIT-licensed. The hosted service and research have their own terms and evidence.

Source: README.md at commit 3b6c420

Tools

0
SourceWeft started this package once in an isolated sandbox, without credentials, to list its tools. That failed: the package type cannot run in the sandbox.

Version history

3
  1. v0.3.2LatestOct 21, 2026
  2. v0.3.1Oct 21, 2026
  3. v0.2.0Sep 16, 2026

Score

50% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts