AutoSearch MCP Server — PRISMA systematic reviews for Claude, ChatGPT, Gemini and Cursor

Why MCP + AutoSearch

Most research MCP servers are thin wrappers around a single search API: they return links, and your agent still has to screen, verify, and synthesize. AutoSearch exposes a complete PRISMA 2020 systematic review pipeline as tools — no commercial research assistant (Elicit, Consensus, Scite, Undermind) offers this over MCP.

Full pipeline, one tool call: question → multi-source search → screening with recorded decisions → evidence synthesis → IMRAD paper with APA references and PRISMA flow counts.
Verified citations by construction: every DOI is checked live against Crossref before it enters a deliverable — and we publish a re-runnable citation-integrity benchmark, failures included.
Async by design: launch a deep review, keep working, poll status, fetch the finished paper as markdown, JSON, or PDF.
8 scholarly sources in one place: OpenAlex, Crossref, PubMed, Semantic Scholar, arXiv, ClinicalTrials.gov, DBLP, DOAJ.

99/99 DOIs resolved (100.0%), 0 fabricated, across 23 real deliverables — See the published benchmark on /trust

Listed on: MCP Registry · Smithery · mcp.so

Get started in 5 minutes

1. Create a free account

Register at exatoshi.uk — you get 50 free credits, no card required. Enough for ~6 standard research runs or 50 evidence searches.

exatoshi.uk/register

2. Generate an API key

Go to Account → API & Integrations, name the key after your client (e.g. "Claude Desktop") and click "Generate new API key". The key starts with as_ and is shown only once — copy it immediately.

Open account API keys

3. Add the server to your client

One command in Claude Code, or a JSON snippet in Claude Desktop / Cursor / ChatGPT (see below).

4. Verify the connection

Run the curl test below: a tools/list response with 6 tools means you are connected.

MCP Server

URL: https://exatoshi.uk/mcp (JSON-RPC 2.0 over HTTP POST)

Manifest: https://exatoshi.uk/.well-known/mcp · alias /.well-known/mcp.json · server card /.well-known/mcp/server-card.json

Auth: x-api-key: as_... — Open account API keys

Install in Claude Code (one command)

claude mcp add --transport http autosearch https://exatoshi.uk/mcp --header "x-api-key: as_YOUR_KEY"

Install in Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) · %APPDATA%/Claude/claude_desktop_config.json (Windows)

{
  "mcpServers": {
    "autosearch": {
      "type": "http",
      "url": "https://exatoshi.uk/mcp",
      "headers": {
        "x-api-key": "as_YOUR_KEY"
      }
    }
  }
}

Restart Claude. autosearch will appear in the tool list.

Install in ChatGPT (Custom Connectors)

In ChatGPT Settings -> Apps & Connectors -> Add new MCP server. Use the URL below and send the API key in the x-api-key header.

URL: https://exatoshi.uk/mcp

Header: x-api-key

Install in Cursor

.cursor/mcp.json

{
  "mcpServers": {
    "autosearch": {
      "url": "https://exatoshi.uk/mcp",
      "headers": {
        "x-api-key": "as_YOUR_KEY"
      }
    }
  }
}

Test with curl

curl -s https://exatoshi.uk/mcp \
  -H "Content-Type: application/json" \
  -H "x-api-key: as_YOUR_KEY" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Tool reference

All calls are JSON-RPC 2.0 POSTs to https://exatoshi.uk/mcp with your key in the x-api-key header. Results arrive as MCP content plus structuredContent.

`autosearch_run_research`Credits: 4 (quick-scan) / 8 (standard-review) / 18 (deep-review)

Launch an AI scientific research run. Returns run_id immediately. Use get_run_status to poll.

Charged once at launch; polling the run afterwards is free.

Example prompts: Run a clinical RAG review on retrieval-augmented generation in hospitals.

Parameters

Parameter	Type	Required	Description	Constraints
`prompt`	string	yes	Research question or topic
`depth`	string	no		one of: quick-scan, standard-review, deep-review; default "standard-review"

Example request

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "autosearch_run_research",
    "arguments": {
      "prompt": "Retrieval-augmented generation for clinical decision support: evidence and limitations",
      "depth": "standard-review"
    }
  }
}

Example response

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "…same payload serialized as JSON text…"
      }
    ],
    "structuredContent": {
      "run_id": "knowledge-lab-rag-clinical-decision-support-20260610-101500",
      "status": "starting",
      "depth": "standard-review",
      "credits_charged": 8,
      "status_url": "https://exatoshi.uk/runs/knowledge-lab-rag-clinical-decision-support-20260610-101500"
    }
  }
}

Common errors

`-32602`	`Missing required param: prompt`	Send a non-empty prompt string.
`-32602`	`Invalid depth`	depth must be quick-scan, standard-review, or deep-review.
`-32002`	`Insufficient credits`	HTTP 402. Top up at /billing or lower the depth.

`autosearch_get_run_status`Credits: free

Get the current status of a research run. Returns status, frontier_score, n_experiments, summary.

Poll as often as you like (rate limits apply).

Example prompts: Check whether my AutoSearch run is ready and summarize the frontier score.

Parameters

Parameter	Type	Required	Description	Constraints
`run_id`	string	yes

Example request

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "autosearch_get_run_status",
    "arguments": {
      "run_id": "knowledge-lab-rag-clinical-decision-support-20260610-101500"
    }
  }
}

Example response

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "…same payload serialized as JSON text…"
      }
    ],
    "structuredContent": {
      "run_id": "knowledge-lab-rag-clinical-decision-support-20260610-101500",
      "status": "completed",
      "frontier_score": 81.1,
      "n_experiments": 6,
      "summary": "Six experiments synthesized; evidence matrix covers 24 peer-reviewed sources.",
      "research_question": "What is the documented evidence on RAG for clinical decision support?",
      "started_at": "2026-06-10T10:15:00Z",
      "last_heartbeat_at": "2026-06-10T10:52:00Z",
      "report_url": "https://exatoshi.uk/runs/knowledge-lab-rag-clinical-decision-support-20260610-101500/report"
    }
  }
}

Common errors

`-32602`	`Missing required param: run_id`	Pass the run_id returned by autosearch_run_research.
`-32602`	`Run not found`	The run_id does not exist or belongs to another account.

`autosearch_get_run_paper`Credits: free

Get the generated paper for a completed run. Returns markdown, JSON metadata, or PDF download URL.

The paper was already paid for by the launch credits.

Example prompts: Fetch the generated markdown paper for this completed run.

Parameters

Parameter	Type	Required	Description	Constraints
`run_id`	string	yes
`format`	string	no		one of: markdown, json, pdf_url; default "markdown"

Example request

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "autosearch_get_run_paper",
    "arguments": {
      "run_id": "knowledge-lab-rag-clinical-decision-support-20260610-101500",
      "format": "pdf_url"
    }
  }
}

Example response

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "…same payload serialized as JSON text…"
      }
    ],
    "structuredContent": {
      "run_id": "knowledge-lab-rag-clinical-decision-support-20260610-101500",
      "format": "pdf_url",
      "url": "https://exatoshi.uk/runs/knowledge-lab-rag-clinical-decision-support-20260610-101500/report.pdf"
    }
  }
}

Common errors

`-32602`	`Invalid format`	format must be markdown, json, or pdf_url.
`-32602`	`Run not found`	Check run_id ownership; papers are private per account.

`autosearch_verify_doi`Credits: free

Verify a DOI via Crossref. Returns validity, title, authors, year, journal.

Live Crossref lookup; no credits charged.

Example prompts: Verify DOI 10.1038/s41586-020-2649-2.

Parameters

Parameter	Type	Required	Description	Constraints
`doi`	string	yes	DOI string with or without https://doi.org/ prefix

Example request

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "autosearch_verify_doi",
    "arguments": {
      "doi": "10.1038/s41586-020-2649-2"
    }
  }
}

Example response

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "…same payload serialized as JSON text…"
      }
    ],
    "structuredContent": {
      "doi": "10.1038/s41586-020-2649-2",
      "valid": true,
      "title": "Array programming with NumPy",
      "authors": [
        "Charles R. Harris",
        "K. Jarrod Millman",
        "Stéfan J. van der Walt"
      ],
      "year": 2020,
      "journal": "Nature",
      "reason": ""
    }
  }
}

Common errors

-32602 Missing required param: doi Accepts bare DOIs or https://doi.org/... URLs.

`autosearch_search_evidence`Credits: 1

Search peer-reviewed evidence across academic sources without launching a full research run. Returns top N papers with metadata.

Flat 1 credit per search, independent of sources or limit.

Example prompts: Find recent papers on quantum dot solar cells.

Parameters

Parameter	Type	Required	Constraints
`query`	string	yes
`sources`	array of string	no	one of: arxiv, clinicaltrials, crossref, dblp, doaj, openalex, pubmed, semantic-scholar; default ["openalex", "crossref", "pubmed", "semantic-scholar"]
`limit`	integer	no	max 50; default 10

Example request

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "autosearch_search_evidence",
    "arguments": {
      "query": "quantum dot solar cells efficiency",
      "sources": [
        "openalex",
        "crossref"
      ],
      "limit": 2
    }
  }
}

Example response

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "…same payload serialized as JSON text…"
      }
    ],
    "structuredContent": {
      "query": "quantum dot solar cells efficiency",
      "sources": [
        "openalex",
        "crossref"
      ],
      "limit": 2,
      "results": [
        {
          "title": "Quantum dot solar cells: progress and prospects",
          "authors": [
            "A. Researcher",
            "B. Scientist"
          ],
          "year": 2024,
          "journal": "Nature Energy",
          "doi": "10.1038/s41560-024-00000-0",
          "url": "https://doi.org/10.1038/s41560-024-00000-0",
          "source": "openalex",
          "snippet": "Recent certified efficiencies above 19% position quantum dot absorbers...",
          "peer_reviewed": true
        }
      ]
    }
  }
}

Common errors

`-32602`	`Missing required param: query`	Send a non-empty query string.
`-32602`	`Invalid sources`	Allowed: openalex, crossref, pubmed, semantic-scholar, arxiv, clinicaltrials, dblp, doaj.
`-32002`	`Insufficient credits`	HTTP 402. Requires at least 1 credit.

`autosearch_list_my_runs`Credits: free

List the authenticated user latest research runs with status and frontier score.

Lists only runs owned by the API key account.

Example prompts: List my latest AutoSearch research runs.

Parameters

Parameter	Type	Required	Description	Constraints
`limit`	integer	no		max 100; default 20

Example request

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "autosearch_list_my_runs",
    "arguments": {
      "limit": 2
    }
  }
}

Example response

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "…same payload serialized as JSON text…"
      }
    ],
    "structuredContent": {
      "runs": [
        {
          "run_id": "knowledge-lab-rag-clinical-decision-support-20260610-101500",
          "status": "completed",
          "frontier_score": 81.1,
          "n_experiments": 6,
          "summary": "Six experiments synthesized; evidence matrix covers 24 peer-reviewed sources.",
          "research_question": "What is the documented evidence on RAG for clinical decision support?",
          "started_at": "2026-06-10T10:15:00Z",
          "last_heartbeat_at": "2026-06-10T10:52:00Z",
          "report_url": "https://exatoshi.uk/runs/knowledge-lab-rag-clinical-decision-support-20260610-101500/report"
        }
      ]
    }
  }
}

Common errors

-32001 Unauthorized Missing or revoked x-api-key header.

Error codes

Errors follow JSON-RPC 2.0: {"jsonrpc":"2.0","id":1,"error":{"code":…,"message":…}}.

`-32001`	Unauthorized	HTTP 401 — missing, invalid, or revoked `x-api-key`.
`-32002`	Insufficient credits	HTTP 402 — top up at /pricing.
`-32029`	Rate limit exceeded	HTTP 429 — back off and retry.
`-32601`	Method not found	Unknown method or tool name.
`-32602`	Invalid params	Missing/invalid argument; message says which.
`-32603`	Internal error	Retry; persistent failures: hello@autosearch.ch.

Rate limits

60 requests/minute per IP and 300 requests/minute per account. Rate-limited calls return JSON-RPC error -32029 with HTTP 429 — back off and retry.

Get an API key

Generate a key in your account, then store it in your MCP client configuration.

Open account API keys

Pricing

Research launches and evidence search consume AutoSearch credits. DOI checks and polling are free.

See pricing

FAQ

How do I get an AutoSearch API key?

Register a free account (50 credits included, no card), then open Account → API & Integrations and click "Generate new API key". Keys are prefixed as_ and shown once at creation; you can revoke them anytime from the same page.

What does a research run cost through MCP?

autosearch_run_research charges 4 credits (quick-scan), 8 (standard-review) or 18 (deep-review) at launch. autosearch_search_evidence costs 1 credit per search. DOI verification, status polling, paper retrieval and run listing are free.

How is AutoSearch different from raw search MCP wrappers?

AutoSearch runs a full PRISMA 2020 systematic review pipeline behind one tool call: multi-source academic search, screening with recorded inclusion/exclusion decisions, evidence synthesis and IMRAD paper generation. Every DOI is verified live against Crossref before it enters the deliverable, and a published citation-integrity benchmark on /trust documents the results — failures included.

Which MCP clients are supported?

Any MCP client that speaks JSON-RPC 2.0 over HTTP with custom headers: Claude Code, Claude Desktop, ChatGPT custom connectors, Cursor, Gemini, and self-built agents. Authentication is a single x-api-key header.

Use AutoSearch from ChatGPT, Claude, Gemini

Why MCP + AutoSearch

Get started in 5 minutes

1. Create a free account

2. Generate an API key

3. Add the server to your client

4. Verify the connection

MCP Server

Install in Claude Code (one command)

Install in Claude Desktop

Install in ChatGPT (Custom Connectors)

Install in Cursor

Test with curl

Tool reference

autosearch_run_researchCredits: 4 (quick-scan) / 8 (standard-review) / 18 (deep-review)

Parameters

Example request

Example response

Common errors

autosearch_get_run_statusCredits: free

Parameters

Example request

Example response

Common errors

autosearch_get_run_paperCredits: free

Parameters

Example request

Example response

Common errors

autosearch_verify_doiCredits: free

Parameters

Example request

Example response

Common errors

autosearch_search_evidenceCredits: 1

Parameters

Example request

Example response

Common errors

autosearch_list_my_runsCredits: free

Parameters

Example request

Example response

Common errors

Error codes

Rate limits

Get an API key

Pricing

FAQ

How do I get an AutoSearch API key?

What does a research run cost through MCP?

How is AutoSearch different from raw search MCP wrappers?

Which MCP clients are supported?

`autosearch_run_research`Credits: 4 (quick-scan) / 8 (standard-review) / 18 (deep-review)

`autosearch_get_run_status`Credits: free

`autosearch_get_run_paper`Credits: free

`autosearch_verify_doi`Credits: free

`autosearch_search_evidence`Credits: 1

`autosearch_list_my_runs`Credits: free