Toolbox

npx add-mcp 'https://toolbox.allthepossibles.com/mcp'

Small tools for people shipping with agents.

Each one does a single mundane job well, runs as a form, a JSON endpoint and an MCP tool, and never calls a model. Use them free. Pay $5 a month if you want them without limits.

Works with Claude, Cursor, VS Code, Codex, ChatGPT, Gemini and Copilot. How to connect.

Search all 14 by the job they do

01 Tells

Find the AI-writing tells in a draft, with their positions, and get a cleaned copy back..

You get: A 0 to 100 score, each tell with its excerpt and the rule behind it, and a cleaned version of the text.

Better than by hand: Finds invisible characters and counts patterns across the whole text, which a read-through by eye does not.

Rule-based pattern checks, not a detector model: a clean score does not prove a person wrote it. Free tier reads 3,000 characters. Free: 20 runs a day, 3,000 chars each. API: POST /api/tells. MCP: check_tells.

See a real before and after

02 Manifest cost

Measure what an MCP server's tool list costs in tokens each session and name the tools that cost most..

You get: Tokens per session start, a per-tool estimate, the three most expensive tools and flags on oversized descriptions.

Better than by hand: Gives a per-tool figure before you add a server, so the choice does not rest on a guess.

The figure is an estimate from character counts (stated in the output), not the exact count from a model's tokenizer. Free: 10 runs a day. API: POST /api/manifest-cost. MCP: manifest_cost.

See a real before and after

03 Agent file lint

Lint a CLAUDE.md, AGENTS.md or MEMORY.md for duplicate rules, contradictions and lines past the 200-line cap..

You get: Findings ranked by severity with line numbers, plus the file's line count and estimated tokens reloaded each session.

Better than by hand: Cross-checks every rule against every other, which is slow to do by eye in a file that has grown over months.

Finds structure and wording problems; it cannot tell whether a rule is right for your project. Free: 20 runs a day. API: POST /api/agent-md-lint. MCP: lint_agent_md.

See a real before and after

04 Install snippets

Turn an MCP endpoint into ready-to-paste install snippets for five apps and confirm the server answers..

You get: Snippets for Claude Code, Claude Desktop, Cursor, Codex and VS Code, plus a live handshake result with tool count.

Better than by hand: One snippet per app in each app's own format, checked against the live server, instead of five separate docs pages.

Probes Streamable HTTP servers only; stdio servers and endpoints that need an interactive sign-in cannot be tested. Free: 30 runs a day. API: POST /api/mcp-install. MCP: mcp_install_snippets.

See a real before and after

05 AI crawlers

See which AI crawlers your robots.txt lets in or blocks, and get a draft llms.txt for the site..

You get: Allowed, blocked or partial for nineteen named crawlers, whether llms.txt exists, and a drafted llms.txt.

Better than by hand: Applies the real robots.txt matching rules (longest user-agent group, then the wildcard), which are easy to misread by hand.

Reads what robots.txt says; it cannot tell whether a crawler obeys it, and blocks at the firewall are invisible to it. Free: 10 runs a day. API: POST /api/ai-crawlers. MCP: check_ai_crawlers.

See a real before and after

07 Page meta

Read one page the way a crawler and a link unfurler do and preview the share card it produces..

You get: Title and description with lengths, canonical, Open Graph, Twitter card, favicon, JSON-LD and the resulting preview.

Better than by hand: Shows every tag a crawler reads on one screen, with the length limits applied, in place of opening view-source and counting.

Reads the HTML the server sends; tags added later by JavaScript are not seen, and it does not render the preview image. Free: 20 runs a day. API: POST /api/meta. MCP: check_page_meta.

See a real before and after

08 JSON to schema

Turn a JSON sample into a JSON Schema and a TypeScript type, with optional keys and formats inferred..

You get: A 2020-12 JSON Schema, a matching TypeScript interface, and a list of every guess made.

Saves about 10 min (our estimate, vs writing the schema and the TypeScript type by hand for an object of about six keys).

Infers from the samples you give it; a key absent from every sample is unknown to it. Free tier reads 20,000 characters. Free: 30 runs a day, 20,000 chars each. API: POST /api/json-schema. MCP: json_to_schema.

See a real before and after

09 Cron explainer

Explain a cron expression in plain English, list its next runs in UTC and warn where schedulers disagree..

You get: A plain-English sentence, the next runs in UTC and warnings for shapes that Workers, GitHub Actions and crontab read differently.

Saves about 3 min (our estimate, vs reading the five fields against a cron reference and working out the next dates by hand).

Five-field expressions only, and every time is shown in UTC; it does not know your scheduler's time zone settings. Free: 30 runs a day. API: POST /api/cron. MCP: explain_cron.

See a real before and after

10 Env scrubber

Turn a.env into a safe.env.example and list the secrets it found by kind and line without echoing them..

You get: A committable.env.example keeping each value's shape, a secrets list with line numbers, and duplicate or quoting findings.

Saves about 5 min (our estimate, vs blanking each value of a typical 10 to 20 line.env by hand and checking for secrets you missed).

Recognizes secrets by key name and known formats; an unusual secret in a plain-looking variable can slip through. Check the result before you commit it. Free: 30 runs a day, 20,000 chars each. API: POST /api/env-example. MCP: scrub_env.

See a real before and after

11 Redirect tracer

Follow a URL hop by hop and flag redirect chains, loops, HTTPS downgrades and dropped query strings..

You get: Every hop with status, destination and timing, then findings on what is wrong with the chain.

Saves about 5 min (our estimate, vs running curl -I by hand for each hop and comparing the results).

Follows up to 10 hops with an 8 second timeout each, and sees only what a plain request sees, not redirects done in JavaScript. Free: 20 runs a day. API: POST /api/redirects. MCP: trace_redirects.

See a real before and after

12 Sitemap check

Read a site's sitemaps, count the URLs and fetch a sample to find the ones that are not a clean 200..

You get: Sitemaps found, URL and host counts, lastmod summary, and the status of each sampled URL.

Saves about 10 min (our estimate, vs opening ten URLs from the sitemap one at a time and noting each status).

Free tier samples up to 25 URLs and reads index files one level deep; it checks status codes, not page quality. Free: 10 runs a day. API: POST /api/sitemap. MCP: check_sitemap.

See a real before and after

13 JWT decoder

Decode a JWT, see when it expires in plain words and catch alg none and sensitive claims..

You get: Header, payload, timestamps in plain dates, expiry in words, and findings on risky settings.

Saves about 3 min (our estimate, vs base64url-decoding the header and payload separately and converting the timestamps by hand).

Decodes only: it never verifies a signature, so a clean result says nothing about whether the token is genuine. Free: 30 runs a day. API: POST /api/jwt. MCP: decode_jwt.

See a real before and after

14 Tool lint

Lint an MCP server's tools for names the API rejects, missing descriptions and unlabeled destructive tools..

You get: Findings per tool, ranked high, medium and low, with the field to fix.

Better than by hand: Reads the definitions the way a model does and catches the names and schemas that make a client refuse the server.

Static checks on the listed definitions; it does not call the tools or judge whether they behave as described. Free: 20 runs a day. API: POST /api/mcp-lint. MCP: lint_mcp_tools.

See a real before and after

Free

Every tool, right here or over the API and MCP, with a daily cap per tool per IP (10 to 30 runs, listed above). Enough to check a post, lint a file, or size up a server before you add it.

Pro $5/mo

• No daily caps on any tool, 300 requests a minute.
• Tells, the linter and JSON to schema take 200,000 characters instead of 3,000 or 20,000.
• One API key that works on every route and on /mcp.
• New tools land here first.

Or take the membership: Pro on this and four other products for $15 a month.