Capture a DOM snapshot of the active page or a targeted subtree

Find elements by CSS selectors or visible text

Run JavaScript in page context and return the result

Query recent HTTP requests from the extension's ring buffers, filtered by tab, platform, status, method, or URL pattern

Get recent failed HTTP requests (4xx/5xx) and console errors

Query WebSocket frame history by platform

Check whether credentials for a social platform are still valid (cookies, headers, tokens)

Check if a tab is showing a rate limit wall, login prompt, captcha, or interstitial

Capture a PNG screenshot of the visible tab area

Get the agent session health snapshot — capture rate, stall detection, auth health, buffer size

Connection health — is the extension connected? Dev mode on? What tabs are open?

List open Chrome tabs with platform detection and capture counts

Get recent console logs filtered by tab, level, and time

Export captured network traffic as HAR 1.2 JSON

Click an element by CSS selector or visible text

Type text into an input, textarea, or contenteditable element

Navigate a tab to a URL (http/https only)

Open a new tab, optionally to a URL

Reload a tab

Scroll a page by pixels (positive = down) or scroll a CSS selector into view. Returns scroll position and whether the page reached the bottom.

Execute a sequence of browser actions (click, type, navigate, wait, evaluate)

Trigger the IG post runner to claim and execute the next pending post task from the Neuron backend. Posts to Instagram using the logged-in session in Chrome.

List all saved workflow recordings

Get the current workflow engine status (idle, recording, replaying, demo)

Start recording user interactions (clicks, typing, navigation) on the active tab as a replayable workflow

Stop recording and save the workflow with a name

Replay a saved workflow — re-executes the recorded steps on the current tab

Stop the current workflow replay or demo

Start a demo video recording of a workflow — replays the steps with cinematic overlays (chapters, captions, cursor) and captures to video

List all active network mock rules

Add a network mock rule — intercept requests matching a URL pattern and return a custom response. Use to test error states, slow APIs, or custom payloads without touching the real backend.

Remove all network mock rules

Run a WCAG accessibility audit on a page — checks color contrast, missing alt text, form labels, heading structure, lang attribute, empty links, and tabindex misuse. Returns structured violations with severity and selectors.

Capture a performance snapshot — Core Web Vitals (LCP, CLS), TTFB, FCP, resource breakdown by type, slow resources, blocking scripts, memory usage, and long tasks.

Get all cookies for a URL, including values, domain, expiry, secure, httpOnly, and sameSite flags

Set a cookie on a domain

Delete a specific cookie by name and URL

Read all localStorage or sessionStorage entries for the current page (values truncated at 500 chars)

Clear all localStorage or sessionStorage entries for the current page

Search across all captured response bodies in ring buffers for a string or regex pattern. Returns matching requests with context snippets around the match.

Analyze captured network traffic and produce a structured API map. Groups endpoints by normalized URL (IDs replaced with {id}), shows call counts, statuses, content types, auth presence, avg duration, and sample response shapes.

Comprehensive security audit of a page. Scans for leaked secrets (API keys, tokens, credentials) in inline scripts, localStorage, and response bodies. Checks security headers, CORS config, mixed content, insecure forms, exposed source maps, and password field autocomplete. Returns structured findings with severity levels.

Capture a snapshot of the current page state — DOM structure, element visibility, text content, meta tags, URL. Store it in memory with a label for later comparison. Use with neuron_diff_states to detect changes over time (useful for SPA testing, mutation tracking, or detecting dynamic content updates).

Compare two page snapshots created with neuron_snapshot_state. Returns added, removed, and changed elements, plus URL/title changes. Useful for detecting what changed on a page after an action (button click, form submit, navigation, etc).

Extract structured data from repeating page patterns (product listings, search results, social feeds, tables). If a CSS selector is provided, extracts from those elements. Otherwise auto-detects repeating patterns (articles, cards, list items) and extracts text, links, images, data attributes. Returns up to 200 items.

Re-fire an HTTP request from the extension's service worker context (with the browser's cookies/session). Modify method, headers, or body before sending. Like Postman but using the live browser session. Response bodies truncated at 50KB.

Network waterfall timing breakdown — DNS, TCP, TLS, TTFB, download timing for each resource loaded by the page, ordered by start time. Shows the critical render-blocking path and identifies the slowest resources.

SEO audit of a page — title, meta description, Open Graph, Twitter Cards, canonical URL, heading hierarchy, image alt text, structured data (JSON-LD/microdata), word count, internal/external link ratio, viewport, lang attribute. Returns findings with severity levels.

Start watching a CSS selector for changes (text, visibility, attributes). Polls at a configurable interval and accumulates changes. Use neuron_get_watches to check for changes, neuron_stop_watch to stop.

Get all active element watches and their accumulated changes

Stop watching an element and return the final list of accumulated changes

Deep-read a single URL in one call. Navigates to the page, waits for load, scrolls to trigger lazy content, extracts structured data, captures console errors, and takes a screenshot. Returns everything the AI needs to understand the page without multiple round-trips. Use instead of separate navigate → scroll → extract → screenshot chains.

Research multiple URLs in parallel. Opens each URL in a separate tab, scrolls to load lazy content, extracts structured data from each, and returns all results in one response. Up to 8 URLs per call. 5-10x faster than researching one at a time. Use for batch profile research, multi-page data collection, or comparing pages side by side.

Run a search on any site and collect results across multiple pages. Navigates to the URL, types the query into the search box, extracts results, optionally paginates (clicks 'next' / scrolls for infinite scroll), and returns all collected items. One call replaces the typical navigate → find search box → type → extract → scroll → extract → click next → extract chain.

Fill a form and optionally submit it in one call. Takes a map of field selectors to values, fills each one, optionally screenshots the filled form for review, and clicks the submit button if specified. Returns the before/after page state diff so you can verify what changed. Replaces the typical find → type → find → type → screenshot → click chain.

Run all quality audits on a page in one call — accessibility (WCAG), security scan, performance snapshot, SEO audit, and console errors. Returns a combined report. Replaces 5 separate tool calls (a11y_audit + security_scan + perf_snapshot + seo_audit + get_errors).

Snapshot page state, perform an action (click/type/navigate), then diff to see what changed. One call replaces the snapshot_state → click → wait → snapshot_state → diff_states chain. Use to verify that a button click, form submit, or navigation actually did what was expected.

Take a screenshot of the page and describe what's visible, then perform an action based on visual understanding — no CSS selectors needed. The extension screenshots the viewport, the agent analyzes the image description, and issues click/type commands using element coordinates or best-match selectors. Use when you don't know the page structure or selectors keep breaking. Describe what you want to interact with in natural language.

Send an approval request to WhatsApp and wait for the response. Takes a screenshot of the current state, sends it with a prompt to a WhatsApp number, waits for approve/reject reply. Use instead of pausing for keyboard approval — the user approves from their phone. Requires a Neuron bot API key and the recipient's phone number.

Extract structured data from a page and return it as clean JSON ready for piping to an API, spreadsheet, or file. Navigates to the URL, extracts data matching a schema you define, and returns normalized rows. Use for building data pipelines — scrape a page and push the data to a webhook, save as CSV, or append to a collection.

Extract and download video/audio from any page. Navigates to the URL, plays the media to trigger network requests, searches captured traffic for video/audio streams (mp4, m3u8, webm, mp3, blob), extracts the CDN URLs, and initiates a browser download. Works on Instagram reels, TikTok videos, X/Twitter videos, Facebook videos, LinkedIn videos, YouTube, and most sites with embedded video. Returns the download URLs found and download status.

Download videos from multiple URLs. Opens each in a tab, extracts video streams, downloads all. Up to 5 URLs per call. Returns results for each URL with download status and any failures. Use for batch downloading from feeds, trending pages, or saved lists.

Start a recipe run. Loads the recipe's agent.md with variables interpolated, its learnings, resolved variables (auto-filled from profile), and active rules (global + platform + task-specific). Rules are injected as non-negotiable constraints that override recipe strategy. Call neuron_recipe_complete when done.

End a recipe run started with neuron_recipe_run. Logs the outcome to the recipe's memory automatically. Provide the outcome data matching the recipe's Reflect section. If the recipe has 5+ memory entries, also return the learnings so the evolve phase can update strategy.

List all available recipes (bundled + user-installed). Shows name, description, whether it has accumulated learnings, and run count.

Get a recipe's full contents — agent instructions (agent.md), configuration (recipe.yaml), accumulated learnings, and variable definitions. Use this to understand what a recipe does before running it.

Create a new recipe from scratch. Provide the slug, name, description, agent instructions (agent.md content with Strategy/Reflect/Evolve sections), and configuration (recipe.yaml content with variables/tools/pipes/limits). Saved to ~/.neuron/recipes//.

Update a recipe's agent instructions, learnings, or configuration. If updating a bundled recipe, it's forked to ~/.neuron/recipes/ first. Use this after the evolve phase to persist improved strategy.

Log a run outcome to a recipe's memory. Each entry captures what happened — success/failure, sources found, dead ends, quality score. The recipe's evolve phase reads these to improve its strategy.

Read a recipe's run history — the outcomes captured by neuron_recipe_log. Returns the most recent entries. Use during the evolve phase to analyze what's working and what's not.

Import a recipe from GitHub or a local path. GitHub: 'github:user/repo' imports all recipes from the repo's recipes/ subdir. Local: absolute path to a folder with recipe.yaml. Saved to ~/.neuron/recipes/. Memory is stripped on import (local-only).

Export a recipe's shareable contents — agent.md, recipe.yaml, and optionally learnings. The recipient imports this with neuron_recipe_import. Memory (run history) is never exported.

Delete a user-installed recipe and all its memory/learnings. Bundled recipes cannot be deleted.

Read the user's profile (~/.neuron/profile.yaml). Profiles auto-fill recipe variables — name, email, location, tone, timezone, etc.

Save or update the user's profile (~/.neuron/profile.yaml). Common fields: name, email, location, timezone, tone, linkedin, github. Recipe variables with matching keys are auto-filled from this profile.

Read the active rules (~/.neuron/rules.yaml). Rules are hard constraints that override recipe strategies. Three levels: 'never' (absolute — never follow anyone, never auto-send without approval), 'global' (apply to every recipe run), 'platform' (per-platform — e.g. linkedin-specific rules). Task-specific rules are passed per-run via neuron_recipe_run.

Set or update rules (~/.neuron/rules.yaml). Rules are non-negotiable constraints that override recipe strategies. Provide any combination of: 'never' (absolute prohibitions), 'global' (apply to all runs), 'platform' (per-platform rules keyed by platform name). Merges with existing rules — platform rules are merged per-platform, global and never are replaced entirely if provided.

First-run onboarding. Checks extension connection status, shows available recipes, checks if a user profile exists, and suggests the best first action. Call this when you first connect to neuron-inspector and aren't sure where to start.

Schedule a recipe to run on an interval or at a specific time

List all scheduled recipes with their next run time, last run time, run count, and enabled status

Remove a scheduled recipe

Enable or disable a schedule without removing it

Save or update session state. Merges with existing state to support resumption across MCP sessions.

Load session state. Returns null if no session exists, allowing recipes to decide whether to resume or start fresh.

List all active sessions with progress summaries. Useful for resuming interrupted work or cleaning up stale sessions.

Delete a session (mark it complete). The session file is removed from disk.

Quick progress update — increment cursor and move an item from working_set to completed. This is the hot-path call that recipes use after each action (e.g. after scraping one profile, sending one message, etc).

Create a persistent page monitor that watches a URL for changes and alerts when conditions are met (price drops, new job postings, content changes). Monitors run in the background at specified intervals.

List all monitors with their status, last check time, and alert count

Remove a monitor by ID

Force an immediate check on a monitor (or all monitors if no ID provided)

Get recent alerts (triggered conditions) from monitors