Your tools, inside the agent.
A Model Context Protocol server that hands your AI coding agent 27 deterministic tools — hashing, encoding, regex, JSON/YAML/SQL formatting, codegen and more. So it stops guessing and starts calling.
- Local stdio via npx
- No telemetry
- 27 curated tools
- Apache-2.0
Add it to your agent in one step
Pick your client and copy the config. It launches locally via npx over stdio — no account, no key, nothing leaves your machine.
Run this once in your terminal — it registers the server globally.
claude mcp add consolelog -- npx -y consolelog-mcpRequires Node.js ≥ 18. The server bundles its own dependencies, so the first npx run is the only setup.
Built for how agents actually work
An agent can already run a CLI. MCP adds discoverability and structured results — and these tools are chosen specifically for where that matters.
Correct, not guessed
Hashes, UUIDs, regex matches and color math are things LLMs get reliably wrong. These tools return deterministic results from the same engine the website runs.
Saves context, not wastes it
Offload bulk formatting, conversion and codegen instead of regenerating output token by token — and the tool set is curated so it never floods the agent with schemas.
One engine, every surface
The exact same logic powers the web app, the CLI and this MCP server, so an agent and a human get identical results.
27 curated tools, not a 300-tool dump
Each one is something an LLM is unreliable at, or that is cheaper to offload than to regenerate. A lean tool set keeps the agent fast and accurate.
Correctness
9Operations language models are reliably wrong at — deterministic by definition.
hashMD5 / SHA-1 / SHA-256 / SHA-512 digest of a string.hmacKeyed HMAC (SHA-256 / SHA-512) of a message.pbkdf2_hashPBKDF2-SHA256 salted password hash (bcrypt-style).uuidRandom RFC-4122 v4 UUIDs.nano_idCompact, URL-safe Nano IDs with chosen alphabet.random_bytesCrypto-random bytes as hex / base64 / binary.passwordStrong random passwords with configurable classes.token_countLLM token count, cost, and context-fit for a model.regex_testActually run a regex and return real matches.
Encoding
7Reversible transforms that get unreliable on longer strings.
base64_encodeUTF-8 string to Base64.base64_decodeBase64 back to UTF-8.url_encodePercent-encode for URLs.url_decodeDecode a percent-encoded string.html_encodeEscape text into HTML entities.html_decodeDecode HTML entities to characters.jwt_decodeDecode a JWT header + payload (unverified).
Format & convert
7Offload bulk rewriting instead of regenerating output token by token.
json_formatPretty-print + validate JSON.json_minifyStrip whitespace from JSON.yaml_to_jsonConvert YAML to JSON.json_to_yamlConvert JSON to YAML.sql_formatPretty-print SQL for a dialect.xml_formatPretty-print + validate XML.csv_to_jsonCSV to a JSON array of objects.
Codegen from data
2Derive types and validators from a real sample — deterministic and tedious.
json_to_typescriptTypeScript interfaces/types from sample JSON.json_to_zodA Zod schema from sample JSON.
Color
2Color-space arithmetic and WCAG math models routinely fumble.
color_convertHEX / RGB / HSL / HSV / CMYK + nearest name.contrast_checkWCAG contrast ratio + pass/fail levels.
Common questions
What it is, which clients work, privacy, and why the tool set is curated.
- What is the consolelog MCP server?
- It is a Model Context Protocol (MCP) server that exposes consolelog.tools utilities directly to AI coding agents. Instead of guessing at a hash, UUID, regex match, or color conversion, an agent calls a tool and gets a correct, deterministic result — backed by the same engine that powers the website and CLI.
- Which AI clients can use it?
- Any MCP-compatible client — Claude Code, Cursor, Claude Desktop, VS Code, Windsurf, and others. The Connect panel above gives you the exact one-line command or JSON config for each.
- Why these tools and not all 300+?
- It is deliberately curated to around two dozen tools. Every tool an MCP server exposes costs the calling agent context on each request, so a bloated server actually makes agents slower and less accurate. We include only what an LLM is unreliable at (hashing, randomness, regex, color math) or what is cheaper to offload than to regenerate (formatting, conversion, codegen), and leave out things the model already does well.
- Does it send my data anywhere?
- No. The default server runs locally over stdio on your machine via npx — your inputs never leave your computer, and there is no telemetry in the local server. It is open source under Apache-2.0.
- Who builds and maintains it?
- consolelog.tools, built and maintained by Matter Development. We build AI tooling, MCP servers, and web products — if you want something like this for your team, get in touch.
Want tooling like this for your team?
consolelog.tools is built by Matter Development — we build AI tooling, MCP servers, and web products. The whole stack here is open source and free to use.