Which integrations are available for this server?

Provides web search capabilities using Brave's independent search index, with a free tier of 2,000 searches per month. Provides AI-powered web search answers using Perplexity's API, with a free tier of $5/month credit.

How do I use captain-search?

1. Click on "Deploy Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@captain-search search for python async await tutorial" That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

captain-search

by mnm-matin

Overview Schema Source Related Servers Discussions Admin

Python

Hybrid

Quick Start

Captain Search should usually start as a CLI workflow, not an MCP setup flow. Install the skill or run the CLI first, prove one command works, and only then add MCP if you explicitly need a server.

From this checkout today

git clone https://github.com/mnm-matin/captain-search.git
cd captain-search
uv sync
uv run csearch skill install --scope user
uv run csearch web "openai api" --max-results 5

After the first PyPI release

uvx --from captain-search csearch skill install
uvx --from captain-search csearch web "openai api" --max-results 5

Related MCP server: Devin Search MCP

Supported Providers

Web Search

You only need one provider to get started. Add more for redundancy.

Provider	Free Tier	Best For	Get API Key
Parallel	$20 on first signup	AI-native web search	platform.parallel.ai
Serper	2,500/month	Google results	serper.dev
Brave	2,000/month	Independent index	brave.com/search/api
Tavily	1,000/month	AI-optimized results	app.tavily.com
Exa	$10 credit	Neural/semantic search	dashboard.exa.ai
Perplexity	$5/mo credit	AI-powered answers	perplexity.ai/settings/api

💡 Tip: Serper + Brave = 4,500 free searches/month with automatic failover

Code Search

Provider	Free Tier	Best For	Get API Key
Exa (MCP)	Unlimited*	Semantic code context	exa.ai
Morph (Warp Grep)	Paid	Repo-local agentic search	morphllm.com
grep.app	Free (no key)	Exact text matching	grep.app
DeepWiki	Free (MCP)	Repo Q&A / Architecture	deepwiki.com
Noodlbox	Temporarily disabled	Local graph analysis	noodlbox.io

Webpage Extraction

Provider	Free Tier	Best For	Get API Key
Parallel Extract	Uses the $20 signup credit	JS-heavy pages, PDFs, clean markdown	platform.parallel.ai
Jina	1M tokens	Webpage/PDF extraction	jina.ai/reader
Trafilatura (fallback)	Free	Local HTML extraction fallback	trafilatura.readthedocs.io
MarkItDown (optional fallback)	Free	Local document → Markdown (DOCX/PPTX/XLSX/...)	github.com/microsoft/markitdown

Enable MarkItDown locally with: uv sync --extra markitdown (minimal) or uv sync --extra markitdown-all (all formats)

Installation

Skill install

If you want an agent to keep reusing Captain Search across sessions or repos, install the Captain Search CLI skill first.

# User-level install from this checkout
uv run csearch skill install --scope user

# Project-local install for the current repo
uv run csearch skill install --scope project

# Claude Code-compatible skill directory
uv run csearch skill install --target claude

# Once published to PyPI
uvx --from captain-search csearch skill install

The installer writes captain-search-cli under .agents/skills by default and can target .claude/skills with --target claude. From a source checkout, --runtime auto bakes in uv run --directory /path/to/captain-search csearch .... From a published package, it defaults to uvx --from captain-search csearch .... Use --force to refresh an existing install.

MCP client install instructions (copy/paste)

Only use this section when the user explicitly wants MCP client setup.

These MCP client configs intentionally use uv run --directory ... because the client may launch the server from an arbitrary working directory and should not depend on your shell having the project virtualenv activated.

You are installing the Captain Search MCP server. First ask the user which API keys they want to configure: SERPER_API_KEY, BRAVE_API_KEY, TAVILY_API_KEY, PERPLEXITY_API_KEY, PARALLEL_API_KEY, EXA_API_KEY, JINA_API_KEY, MORPH_API_KEY. Then ask which client they want to set up (Cursor, Claude Code, VS Code, Windsurf/Antigravity, OpenCode). Use only the keys they choose and remind them at least one web-search key is required.

Then provide the client-specific setup:
- Cursor: give the Captain Search Cursor deeplink and tell them to set the chosen keys as environment variables, then restart Cursor.
- Claude Code: use `claude mcp add captain-search -- uv run --directory /path/to/captain-search csearch mcp` and pass only the chosen keys with `-e KEY=value`.
- VS Code: use the `code --add-mcp` command with only the chosen env keys (or leave them blank if the user wants to fill later).
- Windsurf/Antigravity: add the MCP server to ~/.codeium/windsurf/mcp_config.json with the chosen env keys.
- OpenCode: add to opencode.jsonc under `mcp` with `type: "local"`, `command: ["uv", "run", "--directory", "/path/to/captain-search", "csearch", "mcp"]`, and an `environment` object with the chosen keys.

Install in Cursor

API keys are pulled from your environment (e.g., SERPER_API_KEY). Set them in your shell or system settings, then restart Cursor.

claude mcp add captain-search -e SERPER_API_KEY=your-key -- uv run --directory /path/to/captain-search csearch mcp

Install via terminal:

code --add-mcp '{"name":"captain-search","command":"uv","args":["run","--directory","/path/to/captain-search","csearch","mcp"],"env":{"SERPER_API_KEY":"your-key-here","BRAVE_API_KEY":"","TAVILY_API_KEY":"","PERPLEXITY_API_KEY":"","PARALLEL_API_KEY":"","EXA_API_KEY":"","MORPH_API_KEY":"","JINA_API_KEY":""}}'

Or add to your User Settings (JSON) via Ctrl+Shift+P → Preferences: Open User Settings (JSON):

{
  "mcp": {
    "servers": {
      "captain-search": {
        "command": "uv",
        "args": ["run", "--directory", "/path/to/captain-search", "csearch", "mcp"],
        "env": {
          "SERPER_API_KEY": "your-key-here",
          "BRAVE_API_KEY": "",
          "TAVILY_API_KEY": "",
          "PERPLEXITY_API_KEY": "",
          "PARALLEL_API_KEY": "",
          "EXA_API_KEY": "",
          "MORPH_API_KEY": "",
          "JINA_API_KEY": ""
        }
      }
    }
  }
}

Or add to .vscode/mcp.json in your workspace to share with your team.

Add to your ~/.codeium/windsurf/mcp_config.json (shared config for Windsurf and Antigravity):

{
  "mcpServers": {
    "captain-search": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/captain-search", "csearch", "mcp"],
      "env": {
        "SERPER_API_KEY": "your-key-here",
        "BRAVE_API_KEY": "",
        "TAVILY_API_KEY": "",
        "PERPLEXITY_API_KEY": "",
        "PARALLEL_API_KEY": "",
        "EXA_API_KEY": "",
        "MORPH_API_KEY": "",
        "JINA_API_KEY": ""
      }
    }
  }
}

Add to your opencode.jsonc (project root) under mcp:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "captain_search": {
      "type": "local",
      "command": ["uv", "run", "--directory", "/path/to/captain-search", "csearch", "mcp"],
      "enabled": true,
      "environment": {
        "SERPER_API_KEY": "your-key-here",
        "BRAVE_API_KEY": "",
        "TAVILY_API_KEY": "",
        "PERPLEXITY_API_KEY": "",
        "PARALLEL_API_KEY": "",
        "EXA_API_KEY": "",
        "MORPH_API_KEY": "",
        "JINA_API_KEY": ""
      }
    }
  }
}

git clone https://github.com/mnm-matin/captain-search.git
cd captain-search
uv sync
uv run csearch skill install --scope user

If you want a no-activation, one-off run from this checkout, use uvx:

uvx --from . csearch --help
uvx --from . csearch web "openai api"

That creates console scripts in the project virtualenv. If you activate it, you can run the CLI directly:

source .venv/bin/activate
captain-search --help
csearch --help

If you do not want to activate the virtualenv, you can still call the wrappers directly:

.venv/bin/captain-search --help
.venv/bin/csearch --help

uv run is just the convenience wrapper that runs those same commands inside the project environment.

If you want a persistent command on your shell PATH, install the tool:

uv tool install .

If you publish Captain Search to PyPI, end users can skip cloning entirely:

# One-off runs
uvx --from captain-search csearch --help
uvx --from captain-search csearch web "openai api"

# Persistent install
uv tool install captain-search

Then update /path/to/captain-search in the configs above to your actual path.

Available Tools

Tool	Description
`search_web`	Search with weighted selection and optional multi-provider mode
`search_code`	Search code across Exa, grep.app, DeepWiki, Morph, and local exact matches
`fetch_webpage`	Extract content from any URL (articles, PDFs, docs)

search_web provider selector:

auto (default): weighted single-provider selection with fallback
multi or all: parallel search across all enabled providers
Provider name: serper, brave, tavily, perplexity, parallel, exa, exa_mcp
Comma-separated list for multi-provider search

search_code providers:

Exa Code Context: Semantic search (always runs)
grep.app: Exact text matching (always runs)
DeepWiki: Repo Q&A (requires repo filter)
Morph Warp Grep: Repo-local agentic search (requires repo filter + MORPH_API_KEY)
Local Exact Matches: Repo-local fixed-string search (requires repo filter)
Noodl: Temporarily disabled

CLI Usage

captain-search and csearch expose the same CLI. The docs use csearch, and MCP server startup now requires an explicit mcp subcommand.

# Bare invocation shows help
csearch
csearch version
csearch skill install --scope user

# Explicit MCP server startup
csearch mcp
csearch mcp --transport http --port 8000

# Direct CLI commands
csearch web "openai api" --max-results 5
csearch web "openai api" --all
csearch code "search_web" --repo mnm-matin/captain-search --format json
csearch fetch https://example.com --format json

# No-activation ephemeral runs with uvx
uvx --from . csearch web "openai api"

# Once published to PyPI
uvx --from captain-search csearch web "openai api"

# No-activation form if you want uv to launch the project env for you
uv run csearch web "openai api"

# Module form also works
uv run python -m captain_search web "openai api"

Direct command exit codes are shell-friendly: 0 for successful output, 1 for a top-level command error, and 2 for invalid CLI usage.

Running as a Remote Server

For teams or cloud deployment:

# HTTP mode
captain-search mcp --transport http --port 8000

# SSE mode  
captain-search mcp --transport sse --port 8000

With Authentication

export MCP_AUTH_TOKEN="your-secret-token"
captain-search mcp --transport http --port 8000

Connect via:

Header: Authorization: Bearer your-secret-token
Query: http://host:8000/mcp?token=your-secret-token

Environment Variables

Variable	Required
`SERPER_API_KEY`	At least one provider
`BRAVE_API_KEY`	At least one provider
`TAVILY_API_KEY`	At least one provider
`PERPLEXITY_API_KEY`	Optional
`PARALLEL_API_KEY`	Optional (web search + webpage extraction)
`EXA_API_KEY`	Optional
`MORPH_API_KEY`	Optional (code search)
`MORPH_BASE_URL`	Optional (defaults to https://api.morphllm.com/v1)
`JINA_API_KEY`	Optional (works without)
`MCP_AUTH_TOKEN`	For remote mode
`CAPTAIN_SEARCH_LOG_ENABLED`	Optional (defaults to `true`)
`CAPTAIN_SEARCH_LOG_DIR`	Optional (defaults to `~/.captain-search/logs`)
`CAPTAIN_SEARCH_LOG_FULL_PAYLOADS`	Optional (defaults to `true`)

Every keyed provider also supports a comma-separated *_API_KEYS environment variable for rotation, for example SERPER_API_KEYS, BRAVE_API_KEYS, TAVILY_API_KEYS, PERPLEXITY_API_KEYS, PARALLEL_API_KEYS, EXA_API_KEYS, and JINA_API_KEYS.

Telemetry Logging

Captain Search writes daily JSONL telemetry files to ~/.captain-search/logs/YYYY-MM-DD.jsonl by default.

Each tool call logs:

full input arguments
final output returned to the client
provider attempts and per-provider results
repo resolution and cache events for search_code
Parallel/Jina/fallback stages for fetch_webpage

If you want to reduce what gets stored, set CAPTAIN_SEARCH_LOG_FULL_PAYLOADS=false to log previews instead of full payloads. Set CAPTAIN_SEARCH_LOG_ENABLED=false to disable logging entirely.

How It Works

Weighted Selection: Providers chosen based on free tier limits
Health-Aware Fallback: If one fails or is cooling down, routing skips it and tries the next
Multi-Provider: search_web still supports multi-provider mode internally, while the CLI exposes that breadth as csearch web --all
Fetch Routing: fetch_webpage uses Parallel Extract first when PARALLEL_API_KEY is configured, then Jina Reader, then local fallback
Doctor Command: csearch doctor shows configured providers, recent telemetry, and current cooldown state

Default weights: Serper (42%) → Brave (33%) → Tavily (17%) → Perplexity (8%)

Customize via config.yaml:

providers:
  serper:
    weight: 50
    enabled: true
  brave:
    weight: 30
    enabled: true

Docker

docker build -t captain-search .

docker run -p 8000:8000 \
  -e SERPER_API_KEY=your_key \
  -e MCP_AUTH_TOKEN=your_secret \
  captain-search mcp --transport http --port 8000

Development

git clone https://github.com/mnm-matin/captain-search.git
cd captain-search
uv pip install -e ".[dev]"
pytest

Run the opt-in live provider audit when you want real endpoint checks plus a per-task configured-vs-working summary for web search, code search, and fetch:

RUN_E2E=1 uv run --extra dev python -m pytest -q tests/test_providers.py
LIVE_PROVIDER_CACHE_TTL_SECONDS=900 RUN_E2E=1 uv run --extra dev python -m pytest -q tests/test_providers.py
LIVE_PROVIDER_FORCE_REFRESH=1 RUN_E2E=1 uv run --extra dev python -m pytest -q tests/test_providers.py

The live audit reuses fresh results from .pytest_cache until the TTL expires and prints a session-end summary showing how many active providers are working for each task.

License

MIT License - see LICENSE

Available Tools

3 tools

fetch_webpageFetch WebpageA

Read-onlyIdempotent

Fetch and extract content from a webpage or PDF.

Uses Jina Reader to fetch and convert web pages and PDFs to clean text. Handles JavaScript-rendered pages and extracts content from PDFs. Falls back to Trafilatura extraction if Jina fails.

Args: url: The URL to fetch (web page or PDF)

Returns: Extracted content in markdown format

Examples: - Fetch a webpage: fetch_webpage("https://example.com/article") - Fetch a PDF: fetch_webpage("https://example.com/document.pdf")

Inputs: url (required). Returns: result

search_codeCode SearchA

Read-onlyIdempotent

Search code across multiple providers.

Args: query: Code search query string repo: Git URL or owner/repo (optional). When provided, results are scoped to this repo.

Returns: Search results in markdown format

Inputs: query (required), repo. Returns: result

search_webWeb SearchA

Read-onlyIdempotent

Search the web using weighted selection or multi-provider search.

Args: query: The search query string max_results: Maximum number of results (1-50, default 10). Per provider in multi mode. provider: Provider selector (default: auto)

Returns: Search results in markdown format

Inputs: query (required), provider, max_results. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

Deploy Server

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

TDQS

A4.1/5.0

Scored 2026-08-10 03:10 across 3 tools

Disambiguation5/5

Each tool serves a clearly distinct purpose: web search, code search, and fetching a specific URL. There is no overlap between the tools, so an agent can easily select the correct one.

Naming Consistency5/5

All tool names follow a consistent verb_noun pattern with underscores: search_web, fetch_webpage, search_code. This makes the tool set predictable and easy to navigate.

Tool Count4/5

Three tools is slightly minimal for a search-focused server, but it covers the core operations of web search, code search, and page fetching. The count is reasonable for a niche tool set.

Completeness4/5

The tool set covers the primary search and fetch workflows, but it lacks options like image or news search. Minor gaps exist, but agents can accomplish most tasks without dead ends.

Maintenance

ActivityInactive

ResponsivenessNo recent community issues

Resources

Related MCP Connectors

SearchPipeA
tech.searchpipe
Web search API and MCP server for developers building AI agents.
Search Browser Automation Agent Orchestration
31
https://searchpipe.tech/mcp/
SSearchB
cc.ssearch
Free web search for AI agents. No API key required. Hosted MCP in active development.
Search Browser Automation
82
https://api.ssearch.cc/mcp
MESSORAA
dev.messora
Scrape, crawl and search the web for AI agents via MCP.
Web Scraping Browser Automation Search
1
https://mcp.messora.dev/mcp
agentsearch-mcpA
io.github.IsaiahDupree
LLM-ready web search + instant answers + URL-to-clean-text fetch for agents and RAG.
Search RAG Systems Browser Automation
31
https://agentsearch-mcp.vercel.app/mcp

Related MCP Servers

Web Researcher MCP
Browser Automation Web Scraping Search
zoharbabin
A
license
A
quality
A
maintenance
MCP server for AI-powered web research — search (Google, Brave, Serper, SearXNG), scrape any page, extract PDFs/DOCX/YouTube transcripts, academic & patent search. Single Go binary. Works with Claude, Cursor, Copilot, and any MCP client.
Updated 10 hours ago (2026-11-14 14:47 UTC)
25
77
MIT
Devin Search MCP
Search Web Scraping Documentation Access
soulvon
A
license
B
quality
B
maintenance
Enables AI coding assistants to perform real-time web searches for technical documentation and deep-fetch web pages, returning clean structured results. It supports integration with MCP clients such as Cursor, Claude, and CodeBuddy.
Updated 11 days ago (2026-11-04 12:08 UTC)
3
194 npm
1
MIT
jev-search-mcp
Search Browser Automation
mukiwu
A
license
A
quality
B
maintenance
Enables plain-language web search with automatic source and time-window selection, returning relevance-ranked links and summaries to MCP clients such as Claude Code and Codex. Works as an MCP server, CLI, and Claude Code plugin with zero runtime dependencies.
Updated a month ago (2026-10-02 13:08 UTC)
1
6
MIT
searxNcrawl
Browser Automation Web Scraping Search
Jisung662
F
license
Not graded
quality
A
maintenance
Enables AI agents to search the web through configured providers and crawl individual pages or entire sites into clean Markdown or JSON, including authenticated browsing and deduplicated output. It can run locally over STDIO or HTTP and integrates with MCP clients and command-line workflows.
Updated a month ago (2026-10-17 10:07 UTC)
-

Quick Start

From this checkout today

After the first PyPI release

Supported Providers

Web Search

Code Search

Webpage Extraction

Installation

Skill install

MCP client install instructions (copy/paste)

Available Tools

CLI Usage

Running as a Remote Server

With Authentication

Environment Variables

Telemetry Logging

How It Works

Docker

Development

License

Available Tools

Tool Schema Changelog

TDQS

Maintenance

Resources

Related MCP Connectors

Related MCP Servers

Web Researcher MCP

Devin Search MCP

jev-search-mcp

searxNcrawl