search-boost

Multi-engine web search for coding agents — one SearchBoost core (fused search, fetch, X) with MCP (Cursor, Codex, Claude Code, Grok Build, Antigravity), pi extension, and DeepSeek Harness bundle adapters

Packages

Package details

extension

Install search-boost from npm and Pi will load the resources declared by the package manifest.

npm repo home report

$ pi install npm:search-boost

Package: search-boost
Version: 0.2.2
Published: Sep 22, 2026
Downloads: 130/mo · 130/wk
Author: mr_remon
License: MIT
Types: extension
Size: 1.1 MB
Dependencies: 6 dependencies · 0 peers

Pi manifest JSON

{
  "extensions": [
    "./adapters/pi/index.js"
  ]
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

SearchBoost

Multi-engine web search & evidence synthesis for AI coding agents
One shared core runtime, deeply adapted for MCP, Pi, and DeepSeek Harness

English · 中文文档

[!NOTE] Release & Branch Notice: The formerly standalone pi-search-boost and dsh-search-boost projects are merged into one codebase under lib/. Documented commands specifying @latest retrieve the official published npm package. To test code that is not in a published release, check out the branch or tag you need and follow Installation from Source & Development.

v0.2.2: adds AnySearch, correlation-aware fusion scoring and batched Jev target research with approved-result pagination. Includes release-audit fixes for version matching, publication timestamps and cache clearing. See the release and migration notes. A source merge does not publish npm; @latest only includes this version after a separate npm publication.

Key Features
System Architecture
Supported Hosts Matrix
Quick Start (30 Seconds)
Upgrade & Migration Guide
Interactive Console (TUI)
Tool Suite & Usage Guide
Parallel Multi-Agent Research Workflows
Security
CLI Command Reference (Headless & CI)
Installation from Source & Development
Friendly Links
License

Key Features

Multi-Engine Parallel Fusion (fused_search)
Queries multiple search providers in parallel. Features an out-of-the-box keyless free pool (Bing, DuckDuckGo, Yahoo, Exa-free, AnySearch) and a keyed API pool (Tavily, Brave, Exa, AnySearch). Automatically performs cross-engine URL deduplication, domain routing, and relevance re-ranking.
Clean Webpage Content Extractor (fetch_page)
Fetches the origin first for low latency, with optional same-route curl compatibility fallback and Jina Reader backup. Strips CSS, JS, and ad clutter. Supports focused contextual paragraph extraction via focus, backed by in-memory caching and size limits.
X / Twitter Community Intelligence (x_search)
Retrieves public posts, user timelines, and discussion threads via official xAI API or an anonymous fallback channel. Recovers accurate UTC timestamps from Snowflake post IDs and enforces local author/date filtering without hallucination.
Jev Adaptive Evidence Loop (adaptive_search · Experimental)
Accepts task context, keyword alternatives and acceptance targets. Each round uses at most three Jev calls: plan, score and coverage. Returns approved URLs, titles and descriptions with cursor pagination and no fixed cumulative result-count cap; execution remains budgeted.
Native Multi-Agent Parallel Research
Bundles search-boost and search-boost-parallel-research skills. In hosts supporting subagents (Cursor, Claude Code, Pi, DSH), tasks can be dispatched to parallel Searchers (gathering evidence) and Summarizers (pure synthesis without tools), supporting both Fast and Complex waves.
Unified Core Across All Host Ecosystems
A single, host-neutral core runtime powering standard Model Context Protocol (MCP) servers, alongside native extensions for Pi and Cordis plugin bundles for DeepSeek Harness (DSH).
Zero-Config Onboarding & Strict Security
Requires zero API keys to start using the free engine pool. Sensitive credentials are stored as plaintext in local private configs with restricted file permissions (POSIX 0600); this is not encryption at rest. Uses local networking and proxy DNS, with bounded requests and explicit fallback, with zero credential leakage into model prompts.

System Architecture

SearchBoost follows a "One Core, Three Adapters" architecture. All search logic, content parsing, deduplication algorithms, and network safety mechanisms reside in the shared core:

                    SearchBoost TUI / CLI
             installation · configuration · updates
                              │
                    Shared SearchBoost Core
                    lib/runtime.mjs facade
              search · fetch · X · adaptive evidence
                              │
              ┌───────────────┼───────────────┐
              │               │               │
         MCP Adapter      Pi Adapter     DSH Adapter
         stdio server     extension      Cordis bundle
              │               │               │
     ┌────────┴────────┐      ▼               ▼
     │ Cursor / Claude │      Pi       DeepSeek Harness
     │ Codex / Grok    │
     │ Antigravity     │
     └─────────────────┘

Core (lib/): Host-neutral algorithms, engine orchestration, Jev client protocol, and network safety policies.
Adapters (adapters/): Maps core operations into host-specific protocols (MCP JSON-RPC, Pi Extension API, DSH Cordis lifecycle).
Agents (agents/): Host prompt contracts, workflow templates, and native skill definitions.

Supported Hosts Matrix

Host	Integration Type	Files / Mechanism	Capabilities & Notes
Cursor / Cursor CLI	MCP + Skills	`~/.cursor/mcp.json` / `skills/`	Supports CLI auto-approval (`cli-config.json`), session start injection, parallel research skill
Claude Code	MCP + Prompts	Official `claude` MCP config	Full multi-engine search and page extraction tools with prompt boundaries
Codex	MCP + Prompts	Host MCP configuration	Seamless access to fused search, X retrieval, and reader tools
Grok Build	MCP + Plugin	MCP config + bundled plugin	Automatically syncs and installs companion plugin when `grok` CLI is available
Google Antigravity	MCP + Workspace	Workspace MCP configuration	Per-project workspace guidance and full search tool suite
Pi	Native Extension	`~/.pi/agent/extensions/` / `prompts/`	Registers native extension, `/fast-parallel`, `/complex-parallel`, searcher/summarizer roles
DeepSeek Harness	Native Bundle	`cordis.patch.yml` / `dsh plugin`	Integrated with Cordis runtime; provides native `research_parallel` subagents and call cards

Quick Start (30 Seconds)

Prerequisites

Node.js: >= 22.13.0
Package Manager: npm (or pnpm)

1. Install & Launch the Dashboard

# Install globally
npm install -g search-boost

# Launch the interactive terminal UI (TUI)
search-boost

[!TIP] Zero API Keys Required to Start: SearchBoost includes a robust free engine pool (Bing, DuckDuckGo, Yahoo, Exa-free). You can begin searching immediately without signing up for any paid provider!

2. 3-Step Setup Wizard

Select Setup in the TUI menu. Configure engine keys (or skip to use free tier) and optional X credentials.
Check the agents you want to integrate (Cursor, Claude Code, Pi, etc.), confirming auto-approval and native search replacement.
Restart or reload your chosen agents, then ask them to research anything in conversation!

Upgrade & Migration Guide

1. Routine Updates: Upgrading via TUI

Whether you run unified search-boost or have legacy pi-search-boost / dsh-search-boost installations, select Update in the TUI to upgrade everything:

# Option 1: Open the interactive menu
search-boost
# -> Select "Update"

# Option 2: Run headless update
search-boost upgrade -y

Update Behavior: Fetches the latest published release, updates SearchBoost, and refreshes prompt assets across all configured agents while preserving existing keys, layer choices, and permission settings.

2. Migrating from Legacy `search-boost-mcp`

[!IMPORTANT] If you have the old global package search-boost-mcp installed, npm cannot automatically replace the global binary across package renames. Use the one-line npx migration command:

# Execute safe migration via npx
npx --yes --package=search-boost@latest -- search-boost migrate -y

# Once migrated, routine updates only require:
search-boost
# -> Select "Update" (or search-boost upgrade -y)

Interactive Console (TUI)

Launch search-boost without arguments to access the interactive dashboard built with Clack. Manage installation, updates, and credentials effortlessly:

Menu Option	Purpose
Setup	Complete initial walkthrough: configure engines, search layers, X credentials, and install agents.
Install / update agents	Install or refresh selected host integrations, preserving existing credentials and layer settings.
Update	One-click upgrade: checks npm for updates, upgrades SearchBoost, and refreshes all installed agents (including legacy Pi/DSH adapters).
API keys / Search layer	Manage Tavily, Brave, Exa and optional AnySearch credentials, and change the default search layer (`free` / `api`).
X credentials	Manage X (Twitter) authentication; supports one-click import from local Grok login.
Jev credentials (experimental)	Configure TypeSafe Jev cognitive engine endpoint and Bearer token.
Native web search	Enable or disable host-native search for hosts supporting config switches.
Status	Inspect current engine availability, configuration status, and active integrations.
Print MCP snippet	Print MCP JSON configuration snippets to stdout for manual setups.
Uninstall	Safely remove SearchBoost integrations from selected agents, keeping user configurations intact.

Tool Suite & Usage Guide

When integrated, agents automatically receive standard tool definitions and autonomously determine when to call them.

Tool Responsibilities & Boundaries

Tool	Best Used For	Boundary / Non-Goals
`fused_search`	Parallel multi-engine querying, deduplication, and diversity re-ranking	A single search step; follow-up decisions remain with the parent agent
`fetch_page`	Reading clean content from public URLs with optional keyword focus	Not a browser with login state; cannot access internal/private networks
`x_search`	Retrieving public X posts, author timelines, or discussion threads	Does not guarantee exhaustive comment threads or total sentiment sampling
`adaptive_search`	Experimental: Jev-guided autonomous follow-up and evidence evaluation	Approved URLs and descriptions; model approval is not independent verification
`search_stats`	Reading engine status, memory cache hits, and recent diagnostic stats	Read-only; configuration readiness does not guarantee active external network reachability
`search_layer`	Viewing or switching compatibility search layer in MCP	`show` is read-only; changing layers mutates persistent configuration on disk

1. `fused_search` Multi-Engine Search

Dispatches queries across engines concurrently, normalizes URLs, strips redirects, and applies diversity filters.

Tool Arguments Example (JSON):

{
  "query": "Node.js 22 built-in WebSocket API guide",
  "include_domains": ["nodejs.org", "developer.mozilla.org"],
  "engine_pool": "free",
  "ranking": "balanced",
  "complexity": "simple",
  "max_results": 5
}

engine_pool: free (keyless Bing, DuckDuckGo, Yahoo, Exa-free), api (configured paid engines only), or hybrid (all available engines).
ranking: Scoring presets: balanced (default), research (favors authoritative/documentation sources), or fresh (favors recent publications).
complexity: simple (1 query variant), medium (up to 2 variants), complex (up to 3 deep variants).
community: Boolean (false by default). Set to true to blend real-time X developer discussions into the final result quota.

2. `fetch_page` Smart Content Reader

Reads webpage content from search URLs. Reads and cleans the origin first; curl handles transport compatibility when installed, and Jina Reader is the backup.

Tool Arguments Example:

{
  "url": "https://nodejs.org/api/globals.html",
  "focus": "AbortSignal.any"
}

focus (Optional): Filters and retains paragraphs matching the target keywords.

[!TIP] A focus miss does not prove the information is absent from the page. If in doubt, re-fetch without focus to inspect the full context.

3. `x_search` X (Twitter) Intelligence

Designed for real-time technical tracking and first-party developer updates. Supports keyword search, author timelines, and thread conversations.

Tool Arguments Example:

{
  "query": "Claude 3.7 Sonnet hybrid reasoning from:AnthropicAI",
  "mode": "keyword",
  "max_results": 5
}

Precise Timestamps: When platform timestamps are missing or inconsistent, recovers true UTC creation times from 64-bit Snowflake IDs.
Native Operators: Full support for from:username, since:YYYY-MM-DD, and until:YYYY-MM-DD.

4. `adaptive_search` Jev Evidence Loop (Experimental)

Vercel support: in TUI → Jev credentials, enter https://ai-gateway.vercel.sh/v1 and a Vercel AI Gateway key. SearchBoost selects the official SDK evaluation model typesafe-ai/jev, not chat completions. The default TypeSafe /systemone path remains supported. Both paths use only the canonical user Jev credential, not environment keys, and respect server rate-limit delays.

The caller supplies task context, keyword alternatives and explicit acceptance questions. Jev chooses query strategies and engines, then scores retrieved evidence and judges which targets need another round. Each round has at most three logical requests, not one request per URL.

{
  "tasks": [{
    "context": "ExampleDB 4.2 upgrade impact",
    "targets": [{
      "id": "migration",
      "keywords": ["migration guide", "upgrade guide"],
      "question": "What migration steps are required from 4.1 to 4.2?"
    }]
  }],
  "page_size": 20
}

Up to 6 tasks, 4 targets per task, 12 total targets; legacy questions input remains supported.
results is a flat {url, title, description} list containing only assessed, approved material, not unassessed/rejected candidates.
Read more with {"cursor":"<nextCursor>"}: no new search or Jev call. Pages default to 20 results (max 50) and have a byte budget, but cumulative approved results have no fixed count cap.
Inspect coverageComplete and warnings. End of pagination is not exhaustive search, and model approval is not independent fact verification.
Results are process-local, retained for up to 30 minutes / 32 recent result sets; expiry, eviction or restart invalidates cursors.
Optional task time_range distinguishes publication dates from event dates; unknown dates cannot qualify as today's evidence.

See the adaptive-search contract and limitations.

Engine Pools & Scoring Presets

engine_pool selects engines, ranking selects shared cross-pool weights, and complexity controls only budget. AnySearch is one logical engine: anonymous in free, key-required in api, and key-preferred in hybrid. Configure ANYSEARCH_API_KEY or config keys --set anysearch=KEY.

Engine	balanced	research	fresh
bing	0.957	0.927	1.020
ddg	0.981	0.903	0.927
yahoo	0.957	0.877	0.902
exa-free	1.004	1.085	0.951
tavily	1.049	1.105	1.084
brave	1.004	0.951	1.125
exa	1.049	1.146	1.063
anysearch	1.004	1.042	0.951

These are uncalibrated cold-start priors, not measured quality rankings. consensus-v2.1 combines original ranks, related-provider discounts and max+log consensus, with metadata adjustments capped at 20%. Quality and list-selection scores stay separate; zero-weight sources cannot vote. Recalibrate old min_score thresholds. See scoring and migration and pool routing.

Parallel Multi-Agent Research Workflows

One shared workflow with three host bindings. The parent agent splits a question into independent tracks; each searcher gathers evidence with fused_search and fetch_page; a tool-less summarizer (or the parent itself) then turns the reports into one answer.

                        [ Parent Agent ]
                   splits tracks · owns the answer
                              │
              ┌───────────────┴───────────────┐
              ▼                               ▼
       [ Searcher A ]                  [ Searcher B ]
   fused_search · fetch_page       fused_search · fetch_page
              │                               │
              └───────────────┬───────────────┘
                              ▼
                      [ Summarizer ]
                   no tools · synthesis only
                              ▼
                     [ Parent Report ]

Host	Entry point	How children run
Pi	`/fast-parallel` and `/complex-parallel` prompts; the `search-parallel-subagent` tool; `searcher` / `summarizer` roles	Each searcher is a child process that explicitly loads the SearchBoost Pi extension (`adapters/pi/index.js`) and may call only `fused_search` and `fetch_page`; summarizers launch with `--no-tools`. The caller chooses the wave size — this runner sets no concurrency cap.
DeepSeek Harness	Native `research_parallel` tool	Children spawn on the same Cordis context through the host's subagent service. Searchers receive `toolFilter.allow = ['fused_search','fetch_page']`; summarizers receive an empty allowlist. Before a wave, `research_parallel` verifies both tools are registered and refuses to spawn otherwise. Handles are disposed on success and failure, and `maxDepth` stays at 1, so children cannot nest further research.
MCP hosts (Cursor, Claude Code, Codex, Grok, Antigravity)	Bundled `search-boost-parallel-research` skill	The skill embeds the same roles and workflow and runs them through whatever subagent mechanism the host provides. A host without usable delegation falls back to serial research by the parent.

Installing into an MCP host also installs the companion search-boost routing skill, which picks the right tool for an open-ended question.

Fast mode: one searcher wave, then parent synthesis — no summarizer and no second wave.
Complex mode: one to three waves with a gap review in between; another wave starts only when a material gap remains, and the run stops early when the evidence is sufficient. These wave limits are workflow instructions, not global quotas enforced across independent tool calls.
Fallback: if the host cannot delegate, the parent researches serially and says so instead of pretending a wave ran.

[!IMPORTANT] A child returning ok means it finished with non-empty text, not that its claims are verified. Failed or partial reports stay visible, but their URLs are excluded from the successful aggregate, and final judgment always belongs to the parent.

For Pi/DSH child-tool loading, stale pi-search-boost references, and the retired deep_research tool, see Subagent tool setup and diagnosis.

Security

API keys use a credential store this tool owns. They are not written into prompts, tool results or shell profiles:

Private file permissions: keys live in ~/.search-boost/config/keys.json (redirect the root with SEARCH_BOOST_HOME). On POSIX the store directory is 0700 and the file is 0600; a rewrite is built from a fresh 0600 temporary file rather than written in place, so an existing file is never widened. Backups and upgrade receipts under the same root are 0600 as well. An env-overridden directory keeps its own mode, while the credential file is still created 0600. Windows has no POSIX mode bits — ACL inheritance applies there.
Atomic, locked writes: replacement uses an O_EXCL temporary file plus rename, so a reader sees either the old or the new file and never a partial one; a failed write cleans up after itself and leaves the previous file untouched. Read-modify-write cycles take an exclusive lock, so two writers cannot silently lose each other's update, and a rejected change does not touch the file at all.
Never echoed back: a key is only sent to the engine it belongs to (Tavily, Brave, Exa and AnySearch request parameters/headers). Status output, search-boost config keys --show and the doctor report print masked values (abcd****wxyz); error messages are tested not to contain credential material, and Jev evaluation payloads deliberately carry no key, masked key or fingerprint. AnySearch uses anonymous quota in free, requires a key in api, and prefers a configured key in hybrid. Its potentially credential-bearing error envelopes are never echoed or persisted.

CLI Command Reference (Headless & CI)

In addition to the interactive TUI, SearchBoost provides a comprehensive CLI for scripting and automation:

# ----------------- Core & Interactive -----------------
search-boost                                # Launch interactive dashboard (TUI)
search-boost status                         # Print active configuration and host summary
search-boost --help                         # Display full CLI documentation

# ----------------- Headless Installation -----------------
search-boost install -t cursor -y           # Install for Cursor with auto-approval
search-boost install -t claude,codex --keep-native  # Install while preserving host native search
search-boost install -t antigravity --workspace /path/to/project # Target a specific workspace
search-boost install -t pi -y               # Mount Pi extension and prompt templates
search-boost install -t dsh --profile web   # Connect to DeepSeek Harness web profile
search-boost install -t cursor --dry-run    # Preview installation without writing files

# ----------------- Configuration Management -----------------
search-boost config keys                    # Manage API keys from CLI
search-boost config keys --set anysearch=KEY  # Configure the AnySearch key (ANYSEARCH_API_KEY)
search-boost config layer                   # Switch default layer (free / api)
search-boost config x --import-grok         # Import X credentials from local Grok login
search-boost config jev                     # Configure Jev endpoint and token

# ----------------- Diagnostics & Health -----------------
search-boost doctor                         # Run offline diagnostic checks
search-boost doctor --strict                # Strict mode (exits non-zero on warnings)
search-boost doctor --json                  # Output machine-readable JSON report

# ----------------- Uninstall -----------------
search-boost uninstall -t cursor,claude -y  # Remove integrations from selected hosts

Installation from Source & Development

Use this path to contribute to SearchBoost, or to test code that is not in a published release yet. Everywhere else, @latest refers to the published package.

Prerequisites

Node.js >= 22.13.0 (matching package.json)
git and npm
Optional: curl, for the same-route transport-compatibility fallback in fetch_page

Set up a checkout

# 1. Clone the repository
git clone https://github.com/Mr-remon219/search-boost.git
cd search-boost

# 2. Install dependencies exactly as CI does
npm ci

# 3. Regenerate the bundled Grok plugin assets
npm run plugin:sync-grok

The clone lands on the repository's default branch. To test a different branch or tag, check it out before installing dependencies (git checkout BRANCH_OR_TAG).

Run it without a global install

node cli.mjs                  # interactive TUI straight from the checkout
node cli.mjs status           # one-shot status summary
node cli.mjs install -t pi -y # mount the Pi extension from this checkout
node cli.mjs install -t dsh --profile web

node cli.mjs accepts the same commands as the installed search-boost binary. If you want the bare command in your terminal, link the checkout instead of installing from npm:

npm link        # or: npm install -g .
search-boost

After changing adapter or agent assets, re-run the same install command for that host (node cli.mjs install -t HOST) so the host picks up the new files, then restart or reload that host.

Test gates before a PR

Command	What it covers
`npm run check`	Syntax check across the CLI, core, adapters and scripts
`npm run prepublishOnly`	The full offline suite: plugin sync, syntax, CLI, install, doctor, fusion, X, engines, X auth, Jev, key authority, dry-run, network, search routing, adapters, parallel research, upgrade, MCP and smoke
`npm run test:network`	Proxy retries, curl fallback, request bounds and compatibility regressions
`npm run test:adapters`	MCP, Pi and DSH adapter protocol suites plus Pi subagent settings migration/diagnosis
`npm run test:parallel`	Searcher/summarizer contracts, DSH dispatch preflight, cancellation and tool isolation
`npm run test:adaptive`	Jev adaptive evidence loop
`npm run smoke`	MCP JSON-RPC protocol smoke test

These suites run against loopback fixtures and process doubles, so no engine keys are needed; a green npm run prepublishOnly is the bar for a PR.

Friendly Links

LINUX DO

License

This project is licensed under the MIT License.

SearchBoost

Table of Contents

Key Features

System Architecture

Supported Hosts Matrix

Quick Start (30 Seconds)

Prerequisites

1. Install & Launch the Dashboard

2. 3-Step Setup Wizard

Upgrade & Migration Guide

1. Routine Updates: Upgrading via TUI

2. Migrating from Legacy search-boost-mcp

Interactive Console (TUI)

Tool Suite & Usage Guide

Tool Responsibilities & Boundaries

1. fused_search Multi-Engine Search

2. fetch_page Smart Content Reader

3. x_search X (Twitter) Intelligence

4. adaptive_search Jev Evidence Loop (Experimental)

Engine Pools & Scoring Presets

Parallel Multi-Agent Research Workflows

Security

CLI Command Reference (Headless & CI)

Installation from Source & Development

Prerequisites

Set up a checkout

Run it without a global install

Test gates before a PR

Friendly Links

License

2. Migrating from Legacy `search-boost-mcp`

1. `fused_search` Multi-Engine Search

2. `fetch_page` Smart Content Reader

3. `x_search` X (Twitter) Intelligence

4. `adaptive_search` Jev Evidence Loop (Experimental)