Atlas / MCP servers / 0xmariowu / Autosearch

AutosearchBLOCK

Name: Autosearch
Author: 0xmariowu

mcp/0xmariowu/autosearch

MCP server by 0xmariowu · agent infrastructure · agentic search · ai agent · Audited 2026-10-08

⚑ Report / re-scan Source ↗

Open-source deep research for AI agents: 40 channels, 10+ Chinese sources.

Verdict

BLOCK

Grade

Trust score

64 /100

Exposed tools

23 15r · 8w · 0d

Transport

stdio

License

MIT

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

Open-source Deep Research for AI Agents

40 channels, including 10+ Chinese sources. MCP-native. LLM-decoupled. Plug into the agent host you already use.

[](https://github.com/0xmariowu/Autosearch/actions/workflows/ci.yml) [](https://github.com/0xmariowu/Autosearch/releases) [](https://www.npmjs.com/package/autosearch-ai) [](LICENSE) [](https://github.com/0xmariowu/Autosearch) [](https://modelcontextprotocol.io)

Install · Channels · MCP Setup · Examples · Docs · 中文

AutoSearch is open-source deep research infrastructure built for AI agents. Plug Claude Code, Cursor, Cline, GPT-Researcher, LangChain, LlamaIndex, AutoGen, and other hosts into MCP-native access across 40 channels, including 10+ Chinese sources.

The engine returns indexed multi-source results and stays uncoupled from LLM calls, so your agent keeps its own model, prompts, and workflow.

You ask your AI to research something. It answers from training data cutoff —

"Show me this week's LLM papers on arxiv" → can't, no academic database access
"What are people saying about this product on Reddit" → shallow, only surface-level web results
"Find similar open-source projects on GitHub" → weak, general search misses most repos
"Summarize the Twitter discussion on this topic" → blocked, no public API
"Compare opinions on Hacker News vs Chinese tech forums" → two platforms, manual agg

Read from source at commit d19a3680a96cOBSERVED · 2026-10-08

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add autosearch --env ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY} --env AUTOSEARCH_PACKING_REQUIRE_SECRET_SCAN=${AUTOSEARCH_PACKING_REQUIRE_SECRET_SCAN} --env AUTOSEARCH_PROXY_TOKEN=${AUTOSEARCH_PROXY_TOKEN} --env AUTOSEARCH_SECRETS_FILE=${AUTOSEARCH_SECRETS_FILE} -- uvx autosearch

claude-desktop

{
  "mcpServers": {
    "autosearch": {
      "command": "uvx",
      "args": [
        "autosearch"
      ],
      "env": {
        "ANTHROPIC_API_KEY": "${ANTHROPIC_API_KEY}",
        "AUTOSEARCH_PACKING_REQUIRE_SECRET_SCAN": "${AUTOSEARCH_PACKING_REQUIRE_SECRET_SCAN}",
        "AUTOSEARCH_PROXY_TOKEN": "${AUTOSEARCH_PROXY_TOKEN}",
        "AUTOSEARCH_SECRETS_FILE": "${AUTOSEARCH_SECRETS_FILE}"
      }
    }
  }
}

Exposed tools (23)

15 read · 8 write · 0 destructive.

Tool	Risk	Description
`citation_add`	write	Add URL to citation index (idempotent). Returns {index_id, citation_number, url}.
`citation_create`	write	Create a citation index for a research session. Returns {index_id}.
`citation_export`	read	Export citations as Markdown. Returns {index_id, markdown, count}.
`citation_merge`	write	Merge source citation index into target. Returns {merged_count, skipped_duplicates}.
`consolidate_research`	read	Compress accumulated evidence into a compact research brief.
`context_retention_policy`	read	Trim evidence list to fit within token_budget, keeping highest-scored items.
`delegate_subtask`	write	Run a query across multiple channels concurrently.
`doctor`	read	Scan all configured channels and return their health status with fix hints.
`graph_search_plan`	read	Build a DAG from subtasks and return topologically sorted parallel batches.
`health`	read	Return a structured health snapshot for MCP clients.
`list_channels`	read	List all channels with their runtime availability status.
`list_modes`	read	List available search modes with their channel guidance.
`list_skills`	read	List autosearch skills with their frontmatter metadata.
`loop_add_gap`	write	Mark a topic as a coverage gap. Returns {state_id, gaps}.
`loop_get_gaps`	read	Get coverage gaps for this loop. Returns {state_id, gaps}.
`loop_init`	read	Initialize a reflective search loop. Returns {state_id}.
`loop_update`	write	Update loop state with evidence from run_channel. Returns state summary.
`perspective_questioning`	read	Generate n sub-questions covering different viewpoints on a topic.
`recent_signal_fusion`	read	Filter evidence to items published within the last `days` days, newest first.
`run_channel`	write	Run a single autosearch channel and return raw evidence.
`run_clarify`	write	Run the autosearch clarifier on a user query, returning structured output.
`select_channels_tool`	read	Select 3-8 channels using group-first two-stage algorithm.
`trace_harvest`	read	Extract winning query patterns from a run_channel trace.

Trust audit

BLOCKgrade D · trust 64/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	PASS
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	PASS
L3	Class-specific surface	PASS
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (3 observation(s))
Network: declared (7 observation(s))
Shell: declared (5 observation(s))
Dependencies: not all pinned
Secrets in source: found

Findings (25)

HIGHFilesystem / path · fs.credential_store · CWE-22, CWE-59

autosearch/core/transcribe_path_guard.py:21

"*/.ssh/*",

Why it matters. touches a credential store