neuron-inspector

177 browser tools for AI agents. Your AI can't see your browser. This fixes that.

Inspect DOM, automate clicks, run JavaScript, search network traffic, audit security, check accessibility, scan SEO, mock APIs, extract structured data, record demos, scan profiles, score leads, track competitors, monitor mentions, build revenue pipelines — all from Claude Code, Cursor, Windsurf, or any MCP client.

No API keys. No cloud. Runs on localhost.

npx neuron-inspector

Get started

1. Install the Chrome extension

Download from neuron.ng/extension and sideload:

1. Download and extract the zip
2. Open chrome://extensions and enable Developer Mode
3. Click Load unpacked and select the extracted folder
4. The extension icon appears in your toolbar

2. Connect to your AI tool

Claude Code:

claude mcp add neuron-inspector -- npx neuron-inspector

Cursor — add to .cursor/mcp.json:

{
  "mcpServers": {
    "neuron-inspector": {
      "command": "npx",
      "args": ["neuron-inspector"]
    }
  }
}

Windsurf — add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "neuron-inspector": {
      "command": "npx",
      "args": ["neuron-inspector"]
    }
  }
}

3. Use it

Open any page in Chrome. Ask your AI agent to interact with it. Done.

Multi-Profile Setup

Each Chrome profile can connect to a different bridge instance. This enables multi-account workflows where each profile is logged into different social media accounts.

1. Set the bridge port per Chrome profile

Open the extension popup > Settings (gear icon) > enable Developer Mode. Set a unique Bridge Port for each Chrome profile:

• Profile A (personal): port 7377 (default)
• Profile B (business): port 7378
• Profile C (client): port 7379

2. Start a bridge per profile

Each bridge gets its own port (NEURON_BRIDGE_PORT) and its own data directory (NEURON_DATA_DIR) so leads, monitors, sessions, and recipes stay isolated.

# Terminal 1 — personal (default port + default data dir)
npx neuron-inspector

# Terminal 2 — business
NEURON_BRIDGE_PORT=7378 NEURON_DATA_DIR=~/.neuron/letschop npx neuron-inspector

# Terminal 3 — client
NEURON_BRIDGE_PORT=7379 NEURON_DATA_DIR=~/.neuron/dlvh npx neuron-inspector

3. Register each bridge with your AI tool

claude mcp add rasheed-personal -- npx neuron-inspector
claude mcp add letschop -- env NEURON_BRIDGE_PORT=7378 NEURON_DATA_DIR=~/.neuron/letschop npx neuron-inspector
claude mcp add dlvh -- env NEURON_BRIDGE_PORT=7379 NEURON_DATA_DIR=~/.neuron/dlvh npx neuron-inspector

Trying unpublished tools

Set NEURON_BRIDGE_LOCAL to a bridge checkout and npx neuron-inspector runs that checkout's build instead (run npm run build there first). Unset it to go back to the published package. Reconnect the MCP server after changing it.

claude mcp add rasheed-personal -- env NEURON_BRIDGE_LOCAL=~/code/neuron/packages/extension/bridge npx neuron-inspector

Until a version with this switch is on npm, point the server at the checkout directly: node <checkout>/bin/neuron-inspector.js.

Each bridge logs its data directory on startup so you can confirm isolation:

[bridge] Data directory: /Users/you/.neuron/letschop
[bridge] Primary mode — HTTP + WebSocket server on 127.0.0.1:7378

Data layout:

~/.neuron/              ← rasheed-personal (default)
~/.neuron/letschop/     ← letschop leads, monitors, recipes, collections
~/.neuron/dlvh/         ← dlvh leads, monitors, recipes, collections

What you can do

Debug a page

"Why is this page slow?"

The agent calls neuron_perf_snapshot and gets back Core Web Vitals, render-blocking scripts, memory usage, and the heaviest resources.

"Check this page for security issues"

neuron_security_scan finds leaked API keys in inline scripts, missing CSP headers, mixed content, insecure forms, exposed source maps.

"Is this page accessible?"

neuron_a11y_audit runs a WCAG audit — color contrast failures, missing alt text, broken heading hierarchy, unlabeled form fields.

Explore traffic

"What APIs is this SPA calling?"

neuron_discover_apis maps every endpoint from observed traffic — call counts, status codes, auth patterns, response shapes.

"Reverse-engineer this site's API into a collection"

neuron_api_blueprint analyzes traffic, infers schemas, detects auth patterns, and creates a ready-to-use API collection.

Prospect and generate leads

"Scan these 20 Instagram profiles and score them as leads"

neuron_profile_scanner visits each profile, extracts data. neuron_lead_scorer scores them by follower count, bio keywords, website presence.

"Find people asking for delivery services in this hashtag feed"

neuron_demand_scanner scrolls the feed, matches trigger phrases like "looking for", "need a", "who delivers", and enriches each poster's profile.

"Who are my competitor's most engaged followers?"

neuron_competitor_mapper scans their recent posts, ranks commenters by frequency, and categorizes each as potential customer, fellow business, or influencer.

Monitor and track

"Alert me when a 500 hits /api/orders"

neuron_monitor_create with request_match condition — watches the ring buffers for HTTP requests matching a URL pattern and status range. No tab opened.

"Alert when a console error matching PaymentFailed appears"

neuron_monitor_create with console_match condition — watches console logs for matching messages. No tab opened.

"Trigger actions from a Google Sheet edit"

Every monitor gets a webhook URL. Apps Script calls POST http://127.0.0.1:7377/hooks/:monitorId with a JSON payload and the monitor's action chain executes.

"Watch this page for price drops and auto-screenshot when it triggers"

neuron_monitor_create with price_below condition and a screenshot action. The monitor opens the page, checks the price, and runs the action chain if triggered.

"Track who mentions our brand across platforms"

neuron_mention_tracker scans feeds for your brand terms and saves matches to a searchable bank.

Automate social actions

"Like this post and leave a comment"

neuron_quick_action handles per-platform DOM selectors — works across Instagram, X, LinkedIn, TikTok, Facebook.

"Send personalized DMs to this lead list"

neuron_dm_outreach composes messages from a template with merge fields, paces them to avoid detection, and tracks status.

Toolsets

The bridge has about 190 tools, grouped into toolsets. All of them are on by default. Clients that load every tool schema up front pay about 37k tokens for that; showing only core (39 tools) costs about 7.5k. Claude Code loads MCP tool schemas on demand, so there the full list costs little.

| Toolset | What's in it | |---|---| | core | Browser driving, frames, files, screenshots, page errors and requests, kill switch (always on) | | api | API collections, environments, scenarios, the Scenario Runner | | automation | Recipes, rules, schedules, saved automations, form filling | | monitor | Page monitors | | research | Prospecting: profiles, comments, hashtags, competitors, leads, media grabbing, WhatsApp approvals | | session | Save and restore browser sessions | | devtools | Mocks, cookies, storage, HAR, performance, audits, traffic search, recording and replay |

Opting out per profile with NEURON_TOOLSETS:

| Value | Shows | |---|---| | unset or all | everything | | core,api | only those groups | | all,-research,-devtools (or just -research,-devtools) | everything except those | | auto | core plus the groups this profile used in the last 30 days (everything on a new profile) |

claude mcp add scenario-runner -- env NEURON_TOOLSETS=core,api npx neuron-inspector

Letting the agent adjust it with neuron_toolsets. Changes show up straight away, and save: true makes the result the profile's default next time (NEURON_TOOLSETS, when set, still wins):

neuron_toolsets { action: "list" }                                   // groups, sizes and call counts
neuron_toolsets { action: "disable", toolsets: ["research"], save: true }
neuron_toolsets { action: "enable", toolsets: ["api"] }
neuron_toolsets { action: "auto", save: true }                       // follow usage from now on
neuron_toolsets { action: "reset" }                                  // forget the saved choice

With auto, a hidden group the agent turns on and uses is shown again in later sessions; groups unused for 30 days drop off. Calls are counted per data directory in tool-usage.json, the saved choice is in toolsets.json, and neuron_status shows both.

All 177 tools

Browser Primitives (55 tools)

Low-level extension commands forwarded over WebSocket.

Inspect & Debug

| Tool | What it does | |------|-------------| | neuron_query_dom | DOM snapshot of the page or a subtree | | neuron_find_elements | Find elements by CSS selector or visible text | | neuron_evaluate_js | Run JavaScript in page context | | neuron_get_logs | Console logs filtered by level, tab, search | | neuron_screenshot | PNG screenshot of the visible area | | neuron_diagnose | Connection health check | | neuron_list_tabs | Open tabs with platform detection | | neuron_session_diagnostics | Agent session health | | neuron_action_space | Snapshot all interactable controls on a page |

Browse & Automate

| Tool | What it does | |------|-------------| | neuron_click | Click by selector or text | | neuron_type | Type into inputs and contenteditable | | neuron_scroll | Scroll by pixels or into view | | neuron_navigate | Navigate to a URL | | neuron_press_key | Press keyboard key (Enter, Tab, Escape, etc.) | | neuron_focus_tab | Bring a tab to the foreground | | neuron_open_tab | Open a new tab | | neuron_reload | Reload a tab | | neuron_run_sequence | Multi-step automation (click, type, wait, eval) | | neuron_trigger_post | Trigger the IG post runner |

Network Intelligence

| Tool | What it does | |------|-------------| | neuron_get_requests | Query HTTP requests by tab, method, status, URL | | neuron_get_errors | Recent 4xx/5xx and console errors | | neuron_get_ws_frames | WebSocket and SSE frame history | | neuron_export_har | Export as HAR 1.2 | | neuron_search_traffic | Full-text search across response bodies | | neuron_discover_apis | Auto-map API endpoints from traffic | | neuron_replay_request | Re-fire a request with the browser's live session | | neuron_waterfall | DNS/TLS/TTFB timing per resource |

Security & Auth

| Tool | What it does | |------|-------------| | neuron_security_scan | Leaked secrets, missing headers, CORS, mixed content | | neuron_check_auth | Verify platform credentials | | neuron_detect_blocker | Detect rate limits, captchas, login walls |

Quality

| Tool | What it does | |------|-------------| | neuron_a11y_audit | WCAG audit — contrast, labels, headings, tabindex | | neuron_perf_snapshot | Core Web Vitals, resource breakdown, memory | | neuron_seo_audit | Meta, OG, headings, structured data, links |

Network Mocking

| Tool | What it does | |------|-------------| | neuron_set_mock | Intercept requests, return custom responses | | neuron_get_mocks | List active mock rules | | neuron_clear_mocks | Remove all mocks |

Cookies & Storage

| Tool | What it does | |------|-------------| | neuron_get_cookies | Read cookies for a URL | | neuron_set_cookie | Set a cookie | | neuron_delete_cookie | Delete a cookie | | neuron_get_storage | Read localStorage / sessionStorage | | neuron_clear_storage | Clear storage |

Page State

| Tool | What it does | |------|-------------| | neuron_snapshot_state | Capture page state for later diff | | neuron_diff_states | Compare two snapshots | | neuron_watch_element | Watch a selector for changes | | neuron_get_watches | Check accumulated changes | | neuron_stop_watch | Stop watching | | neuron_extract_data | Extract structured data from feeds, tables, listings |

Demo Recording

| Tool | What it does | |------|-------------| | neuron_start_recording | Record interactions as a workflow | | neuron_stop_recording | Save the recording | | neuron_list_workflows | List saved workflows | | neuron_toggle_workflow | Enable or disable a workflow | | neuron_start_replay | Replay a workflow | | neuron_stop_replay | Stop replay | | neuron_start_demo | Record a demo video with chapters + cursor | | neuron_workflow_status | Engine state |

Compound Tools (15 tools)

Bridge-side orchestration that chains multiple primitives in one call.

| Tool | What it does | |------|-------------| | neuron_research_page | Deep-read a URL: navigate, scroll, extract, screenshot | | neuron_research_profiles | Research multiple URLs in parallel (max 8) | | neuron_search_and_collect | Search a site and collect results across pages | | neuron_fill_and_submit | Fill a form and optionally submit it | | neuron_audit_page | Run all quality audits in one call | | neuron_monitor_action | Snapshot, act, diff to verify changes | | neuron_wait_for | Wait for element, text, or URL condition | | neuron_smart_click | Click with retry, fallback chain, verification | | neuron_smart_type | Type with retry and value verification | | neuron_health_check | Check tab readiness (no blockers, no captcha) | | neuron_vision_act | Screenshot + describe + act (no selectors needed) | | neuron_approve_via_whatsapp | Send approval request to WhatsApp, wait for reply | | neuron_extract_to_json | Extract data as clean JSON/CSV/YAML | | neuron_grab_media | Extract and download video/audio from any page | | neuron_grab_media_batch | Download videos from multiple URLs (max 5) |

Business Intelligence Tools (29 tools)

High-level tools for prospecting, lead generation, competitive intelligence, and social automation. All compound tools — they chain extension primitives with no new extension code.

Prospecting

| Tool | What it does | |------|-------------| | neuron_profile_scanner | Batch-scan profile URLs with session resume support | | neuron_hashtag_scout | Scout hashtag feeds, extract posts and authors | | neuron_local_scout | Search location/explore feeds with profile enrichment and filters | | neuron_bio_harvester | Extract all links from profile bios (websites, WhatsApp, Linktree, email) | | neuron_cross_platform_finder | Find a handle across platforms with confidence scores |

Lead Generation

| Tool | What it does | |------|-------------| | neuron_demand_scanner | Find demand signals via trigger phrases, optional recurring monitor | | neuron_post_to_lead | Monitor post comments for trigger keywords, extract leads | | neuron_comment_miner | Extract commenters from a post as a lead list | | neuron_lead_scorer | Score leads by follower count, bio keywords, website, location | | neuron_revenue_tracker | CRM-lite funnel: discovered to closed-won with revenue attribution |

Competitive Intelligence

| Tool | What it does | |------|-------------| | neuron_competitor_mapper | Map a competitor's engaged audience, categorize commenters | | neuron_influencer_vetter | Trust score (0-100) with red flag detection | | neuron_engagement_analyzer | Analyze your own posting performance with recommendations |

Brand & Content

| Tool | What it does | |------|-------------| | neuron_mention_tracker | Track brand mentions across platforms, optional recurring monitor | | neuron_social_proof | Collect testimonials and positive mentions | | neuron_post_analyzer | Extract metrics and classify posts (single or batch) | | neuron_content_pipeline | Repurpose content across platform formats |

Collections & Monitoring

| Tool | What it does | |------|-------------| | neuron_collection_harvest | One-shot deep-scroll + extract + keyword filter on any feed | | neuron_collection_watch | Persistent monitor for new items in a collection | | neuron_opportunity_radar | Aggregated alert feed from all monitors | | neuron_page_watch | Semantic page change monitor with structured diffs |

Social Automation

| Tool | What it does | |------|-------------| | neuron_quick_action | Platform-aware like/comment/follow/DM (per-platform selectors) | | neuron_dm_outreach | Paced DM campaigns with merge templates | | neuron_group_link_scanner | Extract WhatsApp/Telegram group invite links |

Data Extraction

| Tool | What it does | |------|-------------| | neuron_pattern_learner | Learn CSS selectors from example text, store for reuse |

Network Analysis

| Tool | What it does | |------|-------------| | neuron_site_xray | Full network analysis (requests, perf, security, APIs in parallel) | | neuron_traffic_replay | Capture/replay/list HAR traffic with mock rules | | neuron_api_blueprint | Reverse-engineer APIs from traffic into a collection | | neuron_page_weight | Asset breakdown by category with budget violation detection |

Recipe Tools (17 tools)

Manage and run recipe workflows — purpose-built agents that get better with every run.

| Tool | What it does | |------|-------------| | neuron_recipe_run | Start a recipe with variables and rules | | neuron_recipe_complete | End a recipe run and log the outcome | | neuron_recipe_list | List available recipes | | neuron_recipe_get | Get a recipe's full contents | | neuron_recipe_create | Create a new recipe | | neuron_recipe_update | Update a recipe | | neuron_recipe_delete | Delete a recipe | | neuron_recipe_log | Log a run outcome to memory | | neuron_recipe_memory | Read run history | | neuron_recipe_import | Import from GitHub or local path | | neuron_recipe_export | Export a recipe for sharing | | neuron_profile_get | Read user profile | | neuron_profile_save | Save user profile | | neuron_rules_get | Read active rules | | neuron_rules_set | Set or update rules | | neuron_quickstart | First-run onboarding |

Scheduler Tools (4 tools)

| Tool | What it does | |------|-------------| | neuron_schedule_recipe | Schedule a recipe on interval or cron | | neuron_schedule_list | List all scheduled recipes | | neuron_schedule_remove | Remove a schedule | | neuron_schedule_toggle | Enable or disable a schedule |

Session State Tools (5 tools)

| Tool | What it does | |------|-------------| | neuron_session_save | Save session state for resumption | | neuron_session_load | Load session state | | neuron_session_list | List active sessions | | neuron_session_delete | Delete a session | | neuron_session_checkpoint | Quick progress update |

Monitor Tools (8 tools)

| Tool | What it does | |------|-------------| | neuron_monitor_create | Create a monitor (page, traffic, or console). Returns a webhook URL for external triggers | | neuron_monitor_list | List all monitors | | neuron_monitor_get | Get monitor details | | neuron_monitor_update | Update monitor config | | neuron_monitor_remove | Remove a monitor | | neuron_monitor_toggle | Enable/disable a monitor | | neuron_monitor_check | Force an immediate check | | neuron_monitor_alerts | Get recent alerts |

Condition types:

| Type | What it watches | Needs a tab? | |------|----------------|--------------| | content_change | Any change in page content | Yes | | text_appears | Specific text appears on page | Yes | | text_disappears | Specific text disappears from page | Yes | | price_below | Price drops below a threshold | Yes | | new_items | New items in a list/feed | Yes | | selector_change | Specific CSS selector content changes | Yes | | request_match | HTTP requests matching URL pattern, method, status range | No — queries ring buffers | | console_match | Console messages matching search text and log levels | No — queries ring buffers |

Inbound webhook: Every monitor exposes POST /hooks/:monitorId on the bridge's HTTP server. External systems (Google Sheets Apps Script, n8n, Zapier, cron jobs) can POST to this endpoint to trigger the monitor's action chain. The webhook URL is returned on creation and shown in the UI.

Control Tools (4 tools)

| Tool | What it does | |------|-------------| | neuron_stop | Emergency stop all autonomous operations | | neuron_status | Show what's currently running | | neuron_pause_all | Pause all monitors and schedules | | neuron_resume_all | Resume all paused operations |

API Collection Tools (28 tools)

Full Postman-like API collection management persisted to ~/.neuron/collections/.

| Tool | What it does | |------|-------------| | neuron_api_list_collections | List all collections | | neuron_api_get_collection | Get collection structure | | neuron_api_create_collection | Create a collection | | neuron_api_delete_collection | Delete a collection | | neuron_api_update_collection | Update name/description | | neuron_api_set_defaults | Set collection-level defaults | | neuron_api_upsert_endpoint | Add or update an endpoint | | neuron_api_remove_endpoint | Remove an endpoint | | neuron_api_move_endpoint | Move endpoint between folders | | neuron_api_create_folder | Create a folder | | neuron_api_update_folder | Update folder config | | neuron_api_delete_folder | Delete a folder | | neuron_api_list_folders | List folders | | neuron_api_set_env | Create/update an environment | | neuron_api_remove_env | Remove an environment | | neuron_api_list_envs | List environments | | neuron_api_switch_env | Set active environment | | neuron_api_create_scenario | Create a test scenario | | neuron_api_update_scenario | Update a scenario | | neuron_api_delete_scenario | Delete a scenario | | neuron_api_list_scenarios | List scenarios | | neuron_api_run | Run endpoints against an environment | | neuron_api_run_scenario | Run a scenario with variable chaining | | neuron_api_query | Search endpoints across collections | | neuron_api_generate_tests | Suggest test assertions | | neuron_api_export_collection | Export as YAML or Postman JSON | | neuron_api_import_collection | Import from YAML or Postman JSON | | neuron_api_import_file | Import from a file path |

Automation Tools (12 tools)

Browser automation workflows with scheduling and persistence.

Recipes — purpose-built agents

The 177 tools are primitives. A recipe turns them into a purpose-built agent — a QA engineer, a job applicant, a web researcher. Recipes are shareable, and they get better with every run.

Built-in recipes

| Recipe | What it does | |--------|-------------| | Web Researcher | Deep-dives into topics, cross-references sources, produces reports with citations | | Job Applicant | Searches job boards, evaluates fit, writes cover letters, fills forms, tracks outcomes | | QA Engineer | Tests web apps — finds bugs, checks a11y, audits security, builds regression suites |

How recipes work

A recipe is a folder with instructions (agent.md) and config (recipe.yaml). The instructions tell the AI how to use the browser tools for a specific purpose. The config declares variables you fill in (your resume, your target URL, your preferences).

Recipes are alive. Each run captures outcomes to memory/. After enough runs, the agent reviews what worked and updates its own strategy in learnings.md. A job applicant that discovers technical-tone cover letters get 3x more responses will start writing technical-tone cover letters by default.

Build your own

See RECIPE-SPEC.md for the full format.

Data storage

All persistent data lives in ~/.neuron/:

~/.neuron/
  pipeline.yaml            # Revenue tracker leads
  testimonials.yaml        # Social proof bank
  mentions.yaml            # Brand mention history
  extractors.yaml          # Learned extraction patterns
  group-links.yaml         # WhatsApp/Telegram links
  outreach-campaigns.yaml  # DM outreach campaigns
  monitors.yaml            # Monitor configurations
  monitor-alerts.yaml      # Monitor alert history
  collections/             # API collections (YAML)
  sessions/                # Session state files
  replays/                 # HAR traffic captures
  recipes/                 # Recipe definitions
  rules.yaml               # Global rules
  profile.yaml             # User profile

Development

# Build the bridge
cd packages/extension/bridge
npx tsc

# Build the full extension
cd packages/extension
node build.js

# Type-check without emitting
npx tsc --noEmit

Source files

| File | Purpose | |------|---------| | server.ts | MCP server, HTTP + WebSocket bridge, webhook handler, tool registration | | tools.ts | Browser primitive definitions + name mapping | | compound-tools.ts | Compound tools (research, search, fill, audit, media) | | business-tools.ts | 29 business intelligence and developer tools | | recipe-tools.ts | Recipe management + rules + profile | | scheduler.ts | Cron-like recipe scheduling | | session-state.ts | Session state persistence | | monitor.ts | Page, traffic, and console monitoring with alerts and action chains | | killswitch.ts | Emergency stop, pause, resume | | resilience.ts | Retry, wait, health check helpers | | correlation.ts | WebSocket request/response correlation | | collections/ | API collection CRUD + runner | | automations/ | Browser automation workflows |

Requirements

• Node.js 18+
• Chrome or any Chromium browser
• Neuron extension

Links

• Extension download: neuron.ng/extension
• npm: npmjs.com/package/neuron-inspector
• Recipe spec: RECIPE-SPEC.md

License

MIT