Subindex MCP Server
Integrates with Perplexity AI to conduct web searches and gather verified information for research table generation and validation.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Subindex MCP Server generate a table of top AI companies"
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.
Subindex
Verified Research Engine · subindex.ai · Launch App
Subindex generates, validates, and updates research tables by synthesizing hundreds of calls to Perplexity and Claude. Give it a prompt or an existing table and it returns structured, verified answers across an entire research domain — not just a single query, but a complete field of questions answered at once.
What you want to do | How | Live example |
Gather everything — survey a complete research domain at once | Prompt → structured verified table | |
Monitor anything — news, analyst projections, time-sensitive data | Upload or generate → keep current | |
See everywhere — run the same questions across many entities | One table, many subjects |
How to Access
You want to… | Use |
Try it out or iron out your use case | subindex.ai/app — web GUI for table validation and generation |
Fact-check text or documents interactively | subindex.ai/chex — web GUI for reference checks |
Let an AI agent drive a workflow autonomously | MCP server — install once, describe your task in plain English |
One-off automation without writing code | MCP server via Claude Code, Claude Desktop, or any MCP-compatible client |
Run repeatable pipelines or batch jobs | REST API + example scripts |
Integrate into a product or SaaS | REST API directly |
GUI → API: The web GUIs are ideal for exploring and refining your use case. Once you know what you want, the MCP server or REST API is the better path — faster, repeatable, and fully automatable.
Related MCP server: GPT Researcher MCP Server
Table of Contents
Get Your API Key
Get your API key at subindex.ai/account. New accounts receive $20 in free credits.
Download Examples
All scripts require Python 3.10+ and
pip install requests.
Script | Description | Download |
| Shared REST client (required by all examples) | |
| Validate an existing table | |
| Generate a table from a prompt | |
| Re-run validation on a completed job | |
| Fact-check text or documents |
Or clone the full example set:
# Download all examples at once
curl -O https://subindex-storage.s3.amazonaws.com/website_downloads/examples/subindex_client.py \
-O https://subindex-storage.s3.amazonaws.com/website_downloads/examples/01_validate_table.py \
-O https://subindex-storage.s3.amazonaws.com/website_downloads/examples/02_generate_table.py \
-O https://subindex-storage.s3.amazonaws.com/website_downloads/examples/03_update_table.py \
-O https://subindex-storage.s3.amazonaws.com/website_downloads/examples/04_reference_check.py
pip install requests
export SUBINDEX_API_KEY=sbx_live_...Quick Start: MCP
The MCP server lets any AI agent drive the full Subindex workflow autonomously — no scripting required.
Option A — Direct HTTP connection to Railway (recommended for Claude Code)
Connects directly to the hosted Subindex server over HTTP. No local install, no uvx, no package management — just one command.
Claude Code:
claude mcp add subindex \
--transport http \
https://mcp-server-subindex-production.up.railway.app/ \
--header "X-Api-Key: sbx_live_your_key_here"Via config file (.mcp.json in your repo root, or claude_desktop_config.json):
{
"mcpServers": {
"subindex": {
"type": "http",
"url": "https://mcp-server-subindex-production.up.railway.app/",
"headers": {
"X-Api-Key": "sbx_live_your_key_here"
}
}
}
}Why HTTP over uvx? The HTTP connection runs on Railway — always up to date, no local Python environment needed, and no version drift between the package you installed and the live server. Recommended for Claude Code and any project-level config.
Option B — Local install via uvx
Runs the server locally on your machine using uvx. Useful for Claude Desktop or offline/air-gapped environments.
Claude Code:
claude mcp add subindex uvx mcp-server-subindex \
-e SUBINDEX_API_KEY=sbx_live_your_key_hereClaude Desktop — add to claude_desktop_config.json:
macOS:
~/Library/Application Support/Claude/claude_desktop_config.jsonWindows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"subindex": {
"command": "uvx",
"args": ["mcp-server-subindex"],
"env": {
"SUBINDEX_API_KEY": "sbx_live_your_key_here"
}
}
}
}Project config (shared repo) — add .mcp.json to your repo root. Each person uses their own key; no key is committed to the repo:
{
"mcpServers": {
"subindex": {
"command": "uvx",
"args": ["mcp-server-subindex"],
"env": {
"SUBINDEX_API_KEY": "${SUBINDEX_API_KEY}"
}
}
}
}OpenAI Codex CLI — add to your Codex config file (~/.codex/config.toml on macOS/Linux, %USERPROFILE%\.codex\config.toml on Windows):
[mcp_servers.subindex]
command = "uvx"
args = ["mcp-server-subindex"]
[mcp_servers.subindex.env]
SUBINDEX_API_KEY = "sbx_live_your_key_here"Then restart Codex and verify:
codex mcp listOption C — Smithery
Smithery is an MCP registry that works with Claude Code and other MCP-compatible clients including OpenClaw.
Step 1 — Install and log in:
npx -y @smithery/cli@latest login
npx -y @smithery/cli@latest mcp add subindex/subindex --client claude-codeStep 2 — Authenticate with your API key:
Open your MCP client (e.g. Claude Code), go to /mcp, click subindex → Authenticate, and enter your Subindex API key in the Smithery page that opens.
Smithery login is a one-time step. You must log in before adding servers, or authentication will not be set up correctly.
What to Ask Your Agent
Once the MCP server is installed, describe your task in plain English. The agent drives the full workflow, pausing only when your input is genuinely needed.
Validate a table:
"Validate
companies.xlsxusing Subindex. Interview me about what each column means, then run the preview. If the results look good, approve the full validation."
Generate a table:
"Use Subindex to generate a table of the top 20 US hedge funds with columns: fund name, AUM, primary strategy, founding year, and HQ city. Approve the full validation when the preview looks right."
Re-run validation on the same table:
"Re-run update_table on job
session_20260217_103045_abc123to get an updated validation pass."
Fact-check a document:
"Use Subindex to fact-check this analyst report." (paste the text or share the file path)
Workflows
1. Validate an Existing Table
Minimum rows: Subindex is designed for tables with 4 or more data rows. Fewer rows may produce low-quality results.
Full flow: upload → interview → preview → refine → approve → download
upload_file(filename, file_type, file_size [, file_path]) ← omit file_path on HTTP/Railway transport; run returned curl_command instead
→ start_table_validation(session_id, s3_key, filename)
┌── match found (score ≥ 0.85) → [preview auto-queued; response has preview_queued=true + job_id]
└── no match → interview auto-started
→ wait_for_conversation / poll get_conversation
→ send_conversation_reply (if AI asks questions)
→ [interview complete → preview auto-queued]
→ wait_for_job(job_id or session_id) ← blocks until preview_complete
→ [optional] refine_config(conv_id, session_id, instructions)
→ approve_validation(job_id, cost_usd)
→ wait_for_job(job_id) ← blocks until completed
→ get_results(job_id)Key behavior: The preview is always auto-queued — after the interview finishes (
trigger_config_generation=true), or when a config match is found (match_score ≥ 0.85, response includespreview_queued: trueandjob_id). Callwait_for_job(session_id)directly in all cases (see Config reuse).
Upload interview auto-approval: The interview may auto-approve in a single turn. If the conversation response has
user_reply_needed: falseandstatus: approved, proceed towait_for_job(session_id)immediately — no reply is needed, even if the AI's message appears to ask for confirmation.
Skip the interview with instructions (fire-and-forget config generation):
Pass instructions to start_table_validation to bypass the interactive interview. The AI reads the table structure + your instructions and generates a config directly, then auto-triggers the preview — no clarifying questions needed.
start_table_validation(session_id, s3_key, filename,
instructions="This table lists hedge funds. Validate AUM, strategy, and HQ city. Use Bloomberg and SEC filings.")
→ response includes instructions_mode=true
→ wait_for_job(session_id) ← config generation + preview tracked automatically
→ approve_validation(job_id, cost_usd)
→ wait_for_job(job_id)
→ get_results(job_id)Cost gate: Config generation and the 3-row preview are free. Full validation is charged at
approve_validation— you always see the cost estimate atpreview_completebefore anything is billed. If your balance is insufficient,approve_validationreturns aninsufficient_balanceerror with the required amount.
Refine the config before approving by calling refine_config. This adjusts how columns are validated (sources, strictness, interpretation) — it cannot add or remove columns:
refine_config(conversation_id, session_id,
"Use SEC filings as the primary source for revenue. Require exact match for ticker symbols.")A new preview runs automatically after refinement.
Python script: examples/01_validate_table.py
export SUBINDEX_API_KEY=sbx_live_...
python examples/01_validate_table.py companies.xlsx
python examples/01_validate_table.py companies.xlsx --refine "Add LinkedIn URL column"
Fire-and-forget: provide instructions to skip the interview entirely
python examples/01_validate_table.py companies.xlsx \
--instructions "This table lists hedge funds. Validate AUM, strategy, and HQ city."Which mode to use: Always try
file_pathfirst. If the server can read the file (stdio/uvx transport), upload completes in one step. If you get "File not found" (HTTP/Railway transport — server is remote), call again withoutfile_pathto receive a presigned S3 URL and a ready-to-runcurl_command. The two-step path requires shell access to run curl; Claude Desktop users should use the uvx transport sofile_pathworks.
2. Generate a Table from a Prompt
Describe the table you want — rows, columns, scope — and Subindex builds and validates it from scratch. Designed for tables with 4 or more rows.
start_table_maker("Top 20 US biotech companies: name, ticker, market cap, lead drug, phase")
→ wait_for_conversation / poll get_conversation
→ send_conversation_reply (if AI asks clarifying questions)
→ [table builds → preview auto-queued]
→ wait_for_job(session_id) ← spans table-maker + preview phases
→ approve_validation(job_id, cost_usd)
→ wait_for_job(job_id)
→ get_results(job_id)Auto-approve: The agent can auto-approve the preview and proceed to full validation without human intervention. The preview table is included inline in the
preview_completeresponse.
Cost: ~$0.05/cell (standard), up to ~$0.25/cell (advanced). $2 minimum per run.
Skip confirmation with auto_start=True (fire-and-forget generation):
Pass auto_start=True to skip the AI's clarifying questions and structure-confirmation step. The AI generates the table immediately from the message alone. Use when your message fully describes the desired table.
start_table_maker(
"Top 20 US hedge funds: fund name, AUM, primary strategy, founding year, HQ city",
auto_start=True)
→ wait_for_conversation(conversation_id, session_id)
← returns trigger_execution=true on first response (no Q&A)
→ wait_for_job(session_id) ← table building + preview
→ approve_validation(job_id, cost_usd)
→ wait_for_job(job_id)
→ get_results(job_id)Why
wait_for_conversationwithauto_start=True? Even though there is no Q&A,wait_for_conversationis still required — it returnstrigger_execution: truein a single blocking call (no reply needed), signaling that the table-maker has started. Callingwait_for_jobbefore this call returns would be premature, as the table-maker may not have been triggered yet.
Cost gate: Table building and the 3-row preview are free. Full validation is charged at
approve_validation— you always see the cost estimate atpreview_completebefore anything is billed. If your balance is insufficient,approve_validationreturns aninsufficient_balanceerror with the required amount.
Python script: examples/02_generate_table.py
python examples/02_generate_table.py "Top 10 US hedge funds: fund name, AUM, strategy, HQ city"
python examples/02_generate_table.py --prompt-file my_spec.txt
Fire-and-forget: skip clarifying Q&A and generate immediately from the prompt
python examples/02_generate_table.py --auto-start "Top 10 US hedge funds: fund name, AUM, strategy, HQ city"3. Update a Table (Re-run Validation Pass)
Re-run validation on a completed job — no re-upload or manual edits needed. The table iterates automatically, re-validating the same data with the same config to pick up any changes in source data.
If you want to incorporate manual edits to the output file, re-upload the edited file via upload_file + start_table_validation — a matching config will be found automatically (score ≥ 0.85).
update_table(source_job_id) ← re-validates existing enriched output
→ wait_for_job(new_job_id) ← blocks until preview_complete
→ approve_validation(new_job_id, cost_usd)
→ wait_for_job(new_job_id)
→ get_results(new_job_id)Python script: examples/03_update_table.py
python examples/03_update_table.py session_20260217_103045_abc123
python examples/03_update_table.py session_20260217_103045_abc123 --version 24. Fact-Check Text or Documents (Chex)
Submit any text, report, or document. Subindex checks each factual claim against authoritative sources and returns the same output format as standard table validation: an Excel (XLSX) file, an interactive viewer URL, and a metadata JSON.
Minimum claims: Subindex is designed for text with 4 or more factual claims. Fewer claims may produce low-quality results.
start_reference_check(text="...") ← inline text (or auto_approve=True to skip the gate)
or
upload_file(filename, "pdf", file_size [, file_path]) ← upload PDF/document first
→ start_reference_check(s3_key=s3_key)
→ wait_for_job(job_id) ← spans extraction + 3-row preview; stops at preview_complete
→ preview_table (3 validated sample claims) + cost_estimate shown in response
→ approve_validation(job_id, approved_cost_usd=X) ← triggers Phase 2
→ wait_for_job(job_id) ← waits for completed
→ get_results(job_id) ← download_url (XLSX) + interactive_viewer_url + metadata_urlThree-phase flow: Phase 1 (claim extraction, free) runs automatically, then a 3-row preview validates sample claims (free, auto-triggered). Both phases are tracked by a single
wait_for_jobcall that stops atstatus=preview_complete. Reviewpreview_table(3 validated sample claims with support level and citations) andcost_estimate, then callapprove_validationto start Phase 2 (full validation, charged). Passauto_approve=Trueto skip the gate and run straight through tocompleted.
Progress tracking:
get_job_messagesreturns empty during claim extraction (Phase 1). Progress messages are available during the preview validation phase.wait_for_jobhandles both phases automatically.
Output: Excel (XLSX) file with per-claim rows. Support levels: SUPPORTED / PARTIAL / UNSUPPORTED / UNVERIFIABLE. Share interactive_viewer_url with human stakeholders — it renders sources and confidence scores in a clean UI.
Python script: examples/04_reference_check.py | Sample output: sample_outputs/reference_check_output.json
# Fact-check inline text
python examples/04_reference_check.py --text "Bitcoin was created by Satoshi Nakamoto in 2009."
# Fact-check a PDF
python examples/04_reference_check.py --file analyst_report.pdf
# Fact-check multiple documents concatenated
cat doc1.txt doc2.txt | python examples/04_reference_check.py --stdin--stdin: Concatenates all piped content as a single inline text payload. All claims are attributed to the combined document.
Environment Variables
Variable | Description |
| API key from subindex.ai/account. Required. New accounts get $20 free. |
| Override the API base URL (useful for dev/staging environments). |
Direct REST API
All tools in the MCP server are thin wrappers over the REST API. You can call it directly from any language.
Base URL: https://api.subindex.ai/v1
Auth: Authorization: Bearer sbx_live_your_key_here
Response envelope:
{
"success": true,
"data": { ... },
"meta": { "request_id": "...", "timestamp": "..." }
}Python client (minimal)
import os, requests
BASE_URL = "https://api.subindex.ai/v1"
HEADERS = {"Authorization": f"Bearer {os.environ['SUBINDEX_API_KEY']}"}
def api_get(path, **kwargs):
r = requests.get(f"{BASE_URL}{path}", headers=HEADERS, **kwargs)
r.raise_for_status()
return r.json()["data"]
def api_post(path, **kwargs):
r = requests.post(f"{BASE_URL}{path}", headers=HEADERS, **kwargs)
r.raise_for_status()
return r.json()["data"]A full standalone client module is in examples/subindex_client.py.
API Endpoint Reference
Uploads
Method | Path | Description |
|
| Get a presigned S3 URL to upload a file |
|
| Upload file bytes directly to S3 (no auth header) |
|
| Confirm upload; detect config matches; auto-start interview if no match |
Presigned upload request:
{
"filename": "companies.xlsx",
"file_size": 2048000,
"file_type": "excel",
"content_type": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet"
}Content types: excel → .xlsx, csv → .csv, pdf → .pdf
Confirm upload request (optional fields):
{
"session_id": "session_20260305_...",
"s3_key": "results/.../file.xlsx",
"filename": "companies.xlsx",
"instructions": "Validate AUM, strategy, and HQ city. Use Bloomberg and SEC filings as sources.",
"config_id": "session_20260217_103045_abc123_config_v1_..."
}instructions — if provided, bypasses the interactive upload interview. The AI generates the config directly from the table structure + instructions. Response includes instructions_mode: true and conversation_id. Use wait_for_job(session_id) to track progress — do NOT poll the conversation.
config_id — if provided, skips matching and the interview entirely. The specified config is applied immediately and the preview is auto-queued. Response includes preview_queued: true and job_id. Use wait_for_job(job_id) to track progress. The configuration_id for any completed job is returned by GET /jobs/{id}/results under job_info.configuration_id.
Conversations
Method | Path | Description |
|
| Start a Table Maker session with a natural language prompt |
|
| Poll conversation for status / AI messages |
|
| Send a reply to the AI |
|
| Refine the config with natural language instructions |
Table Maker request body:
{
"message": "Top 20 US hedge funds: fund name, AUM, primary strategy, founding year, HQ city",
"auto_start": true
}auto_start — if true, the AI skips clarifying questions and the structure-confirmation step, proceeding directly to table generation. The first get_conversation response will have trigger_execution: true. Use when your message fully describes the desired table.
Jobs
Method | Path | Description |
|
| Get job status and progress |
|
| Fetch live progress messages (paginated by |
|
| Approve full validation — credits charged here |
|
| Fetch download URL, metadata, viewer URL |
|
| Re-validate enriched output after corrections |
|
| Submit text or file for claim verification |
Job status values:
Status | Meaning |
| Accepted, waiting to start |
| Actively running |
| Free preview done — review results and approve full run |
| Full validation complete, results ready |
| Error — check |
Account
Method | Path | Description |
|
| Current credit balance and this-month usage |
|
| Billing history (supports |
MCP Prompts
Three built-in prompts act as workflow starters — select them from the prompt picker in your MCP client (Claude Code: / menu; Claude Desktop: the prompt icon) and fill in the arguments.
Prompt | Arguments | What it does |
|
| Builds a step-by-step instruction for creating a new research table from a natural language description |
|
| Generates the full validation workflow for an existing Excel or CSV file |
|
| Generates the reference-check workflow for fact-checking a text passage |
MCP Tool Reference
Every tool response includes a _guidance block with a plain-English summary and the exact next tool call(s) — enabling fully autonomous agent workflows.
Tool | Description |
| Upload Excel, CSV, or PDF. Try with |
| Confirm upload; detect config matches; auto-start interview if needed |
| Start an AI conversation to generate a table from a prompt |
| Poll a conversation for AI responses or status changes |
| Reply to AI questions during an interview or table-maker session |
| Block until conversation needs input or finishes (emits live progress) |
| Refine the validation config with natural language instructions (adjusts sources, strictness, interpretation — cannot add or remove columns) |
| Block until |
| One-shot status poll |
| Fetch progress messages with native percentages (paginated) |
| Approve preview → start full validation (credits charged here) |
| Download URL, inline metadata, interactive viewer URL |
| Re-validate enriched output after analyst corrections |
| Submit text or file for claim and citation verification |
| Check credit balance |
| Review billing history |
Key Behaviors
Auto-queued preview
The preview is automatically queued in all three paths after start_table_validation:
Path | Trigger | What to call next |
Config match (score ≥ 0.85) |
|
|
|
|
|
Interview ran |
|
|
To reuse a config from a different session, pass config_id to start_table_validation — the preview will be auto-queued immediately.
Config reuse
If start_table_validation returns match_score ≥ 0.85, the preview is automatically queued using the matched config. The response includes preview_queued: true and job_id — call wait_for_job(job_id) directly, no interview needed.
The configuration_id from any completed job's get_results response can be reused on future uploads of similar tables.
Cost confirmation gate
approve_validation requires approved_cost_usd matching the preview estimate. This prevents surprise charges. The estimate is in the preview_complete job status response under cost_estimate.estimated_total_cost_usd.
This gate applies regardless of whether instructions or auto_start was used — both only skip the interview/confirmation conversation, not the cost approval step. If your balance is insufficient when approve_validation is called, the API returns:
{ "error": "insufficient_balance", "required_usd": 4.20, "current_balance_usd": 1.50 }Fire-and-forget shortcuts
Two optional flags let fully automated pipelines skip interactive steps:
Flag | Tool | Skips | Next step |
|
| Upload interview Q&A |
|
|
| Structure confirmation |
|
These flags use different terminal signals: instructions= (a config-gen flow) causes trigger_config_generation: true on the conversation response; auto_start=True (a table-maker flow) causes trigger_execution: true. Both skip interactive Q&A but produce different fields — do not wait for trigger_execution when using the instructions= upload path. The preview_complete cost gate and approve_validation still apply.
Consuming results: humans vs AI agents
Output files generated per run:
File | Format | Description |
Preview table | Markdown (inline) | First 3 rows as markdown text; returned inline in the |
Enriched results | Excel ( | Ideal for sharing with humans; sources and citations are embedded in cell comments |
Full metadata |
| Complete per-cell detail for every row; use the |
get_results returns:
Field | Type | Best for |
| URL | Humans — web viewer with confidence indicators (requires login at subindex.ai with the same email as your API key) |
| Presigned URL | Humans — download the enriched Excel (.xlsx) directly |
| Presigned URL | AI agents — JSON file with all rows, per-cell details, and source citations |
Recommended AI agent workflow:
At
preview_complete: read the inlinepreview_table(markdown, 3 rows) fromGET /jobs/{id}to survey the table structure and spot-check values. The AI agent can review this inline table and callapprove_validationdirectly — no human approval step is required.After full validation: fetch
results.metadata_url→table_metadata.json. This contains every validated row.Use
rows[].row_key(stable SHA-256) to cross-reference rows between the markdown summary and the detailed JSON.Per-cell fields in
table_metadata.json:cells[col].value— validated value (legacy files may usefull_value)cells[col].confidence—HIGH/MEDIUM/LOW/IDcells[col].comment.validator_explanation— reasoningcells[col].comment.key_citation— top authoritative sourcecells[col].comment.sources[]— all sources withurlandsnippet
Pricing
Mode | Cost |
Preview (first 3 rows) | Free |
Standard validation | ~$0.05 / cell |
Advanced validation | up to ~$0.25 / cell |
Minimum per run | $2.00 |
Reference check | TBD — contact support |
Credits are prepaid. Get $20 free at subindex.ai/account.
Standard validation is used for most tables. Advanced validation is selected automatically when the table requires more sophisticated reasoning (e.g., scientific data, complex financial metrics, or cells with high ambiguity).
Links
MCP server (HTTP, recommended):
claude mcp add subindex --transport http https://mcp-server-subindex-production.up.railway.app/ --header "X-Api-Key: sbx_live_..."— no install neededMCP server (PyPI/uvx):
uvx mcp-server-subindex— for Claude Desktop or offline useDocumentation: subindex.ai/mcp
API reference: subindex.ai/api
Account & credits: subindex.ai/account
Available Tools
25 toolsadd_pending_rowA
Add a new entity as a pending row to be included in the full validation run.
The source Excel is NOT modified — the row is stored in session state and injected in-memory at validation time. Fully reversible before approving the full run.
The full-run cost quote updates to include the new row count.
Inputs: session_id (required), entity_id (required), entity_name (required), extra_fields. Returns: result
add_validated_rowsA
Add new rows to a completed validation table.
Deduplication runs against existing rows. If confirmed=False, returns a cost quote (N_new_rows x per_row_rate from last run). If confirmed=True, appends rows to source Excel, runs validation on new rows only, and merges results into the output Excel.
Only available after full validation completes (status=completed).
Inputs: session_id (required), entities (required), confirmed. Returns: result
approve_validationA
Approve a preview and start full validation processing.
job_id: the session_id value — "job_id" and "session_id" are the same string.
approved_cost_usd MUST be provided and must match the estimated cost from the preview_complete response. This prevents accidental billing without first reviewing preview results and the cost estimate.
Workflow:
Call wait_for_job to reach preview_complete — read the inline preview_table (3-row sample) and cost_estimate.estimated_total_cost_usd from the response
Call approve_validation(job_id=<session_id>, approved_cost_usd=)
Inputs: job_id (required), approved_cost_usd. Returns: result
discover_rowsA
Discover and add new rows to an existing validated table using AI-powered search.
Uses the existing table's config and validated data to plan a targeted row discovery run. The planner derives search strategy from the config, then RowDiscovery finds candidates and QC filters them.
If confirmed=False, returns a cost estimate. If confirmed=True, enqueues the discovery pipeline (planner -> search -> QC -> pending_rows).
Discovered rows land in pending_rows with source='row_discover'. Run add_validated_rows to validate them, or trigger a preview to see them.
Inputs: session_id (required), instruction (required), count, confirmed. Returns: result
exclude_rowA
Exclude a row from the full validation run.
Call with confirmed=False first to see a warning, then re-call with confirmed=True to apply. Reversible via include_row at any point before approving the full validation run.
Inputs: session_id (required), row_key (required), confirmed. Returns: result
get_balanceARead-only
Return the current account credit balance in USD.
Inputs: no named parameters. Returns: result
get_conversationARead-only
Poll a conversation for new messages or a status change.
Key statuses: processing → poll again in ~15s user_reply_needed → send_conversation_reply trigger_execution → preview is auto-queued; switch to get_job_status
Inputs: conversation_id (required), session_id (required). Returns: result
get_job_messagesARead-only
Fetch live progress messages for a running job.
Pass since_seq from the last message to receive only new messages.
Inputs: job_id (required), since_seq. Returns: result
get_job_statusARead-only
One-shot job status check. Prefer wait_for_job for tracking long-running jobs.
Use this for quick status inspections or as a fallback when wait_for_job is not appropriate. For polling loops, wait_for_job is more efficient — it holds the connection, emits live MCP progress notifications, and handles the multi-phase pipeline (table-maker → preview) automatically.
Key statuses: queued / processing → call wait_for_job instead of re-polling manually preview_complete → approve_validation (or refine_config) completed → get_results failed → check error field
Inputs: job_id (required). Returns: result
get_preview_stateARead-only
Get the current structural editing state of a session.
Returns the preview table, excluded rows, pending rows, ignored columns, and current row order so you can review before triggering the full validation run.
Inputs: session_id (required). Returns: result
get_resultsARead-only
Fetch the final validated/enriched results for a completed job.
job_id: the session_id value — "job_id" and "session_id" are the same string.
Automatically downloads table_metadata.json and embeds it inline so no separate HTTP fetch is needed. Key fields in the response:
results.markdown_table — START HERE. Self-contained markdown document: full validated table (all rows, all values), confidence icons, viewer/download links, and a guide to navigating citations. Read this first. results.metadata.rows[] — per-row data keyed by row_key; each cell has value, confidence, comment (with citations). (legacy files may use full_value; both are equivalent) results.interactive_viewer_url — share with humans; renders sources + confidence. results.download_url — enriched Excel file for offline sharing.
Inputs: job_id (required). Returns: result
get_usageARead-only
Return API usage history. Dates in YYYY-MM-DD format.
Inputs: start_date, end_date, limit, offset. Returns: result
include_rowA
Re-include a previously excluded row.
No confirmation needed. Can be called at any point before approving the full validation run.
Inputs: session_id (required), row_key (required). Returns: result
patch_columnA
Add a new column to a completed validation table.
If confirmed=False, returns a cost ceiling estimate (max-case: all rows x all columns x per_cell_cost x 1.25; likely much less if run within 1 day of original validation due to cache).
If confirmed=True, adds column header to source Excel, runs validation with updated config (old columns hit cache, new column fully validated), merges new column results into output Excel. No QC on column patch runs (single-column QC is not meaningful; full-table QC ran on the original validation).
Only available after full validation completes (status=completed).
Inputs: session_id (required), column_name (required), validation_target (required), confirmed. Returns: result
refine_configA
Refine the generated validation config using natural language instructions.
Example instructions: 'Add a column for LinkedIn URL. Remove the revenue column. Make email validation stricter.'
Set defer_preview=True if you plan to do structural editing (exclude_row, add_pending_row, etc.) before the preview — this prevents a premature auto-preview from firing before your edits are complete.
Inputs: conversation_id (required), session_id (required), instructions (required), defer_preview. Returns: result
reorder_preview_rowsA
Set the output row order for the validation run.
This is a display preference — it does not affect which rows are validated or the validation itself. Rows not in the list are sorted to the end.
Inputs: session_id (required), row_keys_in_order (required). Returns: result
send_conversation_replyA
Send a user reply in an ongoing conversation (interview or table-maker).
After sending, poll get_conversation for the AI's next response.
Inputs: conversation_id (required), session_id (required), message (required). Returns: result
start_reference_checkA
Submit a reference-check job to fact-check text or a document.
For inline text: start_reference_check(text="The claims to fact-check...")
For a PDF or document, upload it first then pass the s3_key: upload_file(file_path, file_type="pdf") → returns s3_key start_reference_check(s3_key=s3_key) Do NOT call start_table_validation for PDFs — that starts the table validation pipeline, which is not what you want for a reference check.
Designed for text with 4 or more factual claims; fewer claims may produce low-quality results.
Three phases:
Phase 1 (free): claim extraction + 3-row preview validation (auto-triggered). wait_for_job blocks until status=preview_complete. Review preview_table (3 validated sample claims) and cost_estimate.
Approval gate: call approve_validation to proceed.
Phase 2 (charged): full claim validation. Returns XLSX, viewer URL, metadata.
Set auto_approve=True to skip the approval gate and run straight through to completion automatically.
Inputs: text, s3_key, auto_approve. Returns: result
start_table_makerA
Start a Table Maker conversation to generate a research table.
Describe the table you want in natural language, e.g.: 'Create a table of AI startups that raised Series A in 2024 with columns: company name, funding amount, investors, product description.'
auto_start: When True, the AI skips the confirmation step and generates the table immediately from the message alone, without asking clarifying questions or showing a structure for approval. Use when the message fully describes the desired table and no back-and-forth is needed.
Inputs: message (required), auto_start. Returns: result
start_table_validationA
Confirm the upload and detect matching prior configs.
Call this immediately after upload_file completes (or after the curl upload finishes when using HTTP/Railway transport). Returns config_matches with match_score — if score >= 0.85 a prior config can be reused directly.
instructions: Optional natural-language description of what to validate and how (e.g. "This table lists clinical trials — validate that trial IDs, phase, and primary endpoints are accurate"). When provided, the upload interview is bypassed: the AI reads the table structure + instructions and generates a config directly without asking clarifying questions. Preview is auto-triggered immediately after.
config_id: Optional ID of a known prior configuration to reuse directly. When provided, skips matching and the interview entirely — applies the config and queues the preview immediately. Response includes preview_queued=true and job_id. Use when you already know the config_id (e.g. from a previous job's get_results response).
Config generation and the 3-row preview are free. Full validation is charged at approve_validation — you still see the cost at preview_complete before anything is billed. If balance is insufficient at that point, approve_validation returns an insufficient_balance error.
Inputs: session_id (required), s3_key (required), filename (required), instructions, +1 more. Returns: result
trigger_previewA
Trigger a preview run after structural editing is complete.
Call this after finishing row/column edits (exclude_row, add_pending_row, add_column, etc.) to explicitly queue a preview job.
This clears skip_auto_preview and queues the preview. Then call wait_for_job(session_id) to track progress.
Inputs: session_id (required). Returns: result
update_tableA
Re-run validation on a previously processed table (update in place).
source_version can pin a specific prior result version; omit for latest.
Inputs: source_job_id (required), source_version. Returns: result
upload_fileA
Upload a file to Subindex.
RECOMMENDED: always try with file_path first. If the server can read the file (stdio/uvx transport), it uploads in one step and returns session_id
s3_key immediately. If the server is remote (HTTP/Railway transport), file_path raises "File not found" — in that case call again without file_path to get a presigned S3 upload_url and curl_command instead.
With file_path (one step — stdio/uvx transport): upload_file(filename="data.xlsx", file_type="excel", file_size=12345, file_path="/abs/path/data.xlsx") → server reads + uploads → returns session_id, s3_key
Without file_path (two step — HTTP/Railway transport or any remote server): upload_file(filename="data.xlsx", file_type="excel", file_size=12345) → returns upload_url + curl_command → run the curl_command (requires shell/Bash access), then call start_table_validation(session_id, s3_key, filename)
Note: the two-step path requires shell access to run curl. If you have no shell (e.g. Claude Desktop), use the stdio/uvx transport instead so file_path works.
file_type must be one of: "excel", "csv", "pdf" The presigned URL expires in ~15 minutes — run curl immediately.
Inputs: filename (required), file_type (required), file_size (required), file_path, +1 more. Returns: result
wait_for_conversationARead-only
Wait for a conversation turn to complete, emitting live synthetic progress.
Preferred over manually polling get_conversation. Since conversation processing has no native progress signal, this tool emits time-based synthetic progress — advancing quickly at first, then slowing as it approaches expected_seconds — so the MCP host shows a "still thinking" indicator rather than a frozen bar.
Returns when any of these conditions are met: user_reply_needed=True → AI asked a question; call send_conversation_reply trigger_execution=True → AI approved execution; preview is auto-queued, switch to wait_for_job(session_id) Non-processing status → unexpected terminal (inspect status field) Timeout → returns last known state with _wait_timeout note
Applies to all conversation types: upload interview, table-maker interview, config refinement.
expected_seconds: typical AI response time for this turn (default 120). First table-maker turn (research + planning): ~120–180s. Upload interview first turn (CSV analysis + plan): ~90–150s. Follow-up confirmations ("yes, proceed"): ~30–60s. poll_interval: seconds between status checks (default 8). timeout_seconds: max wall time before returning (default 900). Upload interview turns can take up to 15 minutes — set accordingly.
Inputs: conversation_id (required), session_id (required), expected_seconds, timeout_seconds, +1 more. Returns: result
wait_for_jobARead-only
Wait for a job to reach a terminal state, emitting live MCP progress notifications.
Preferred over manually looping get_job_status. The MCP host shows a live progress indicator while this tool holds the connection — no extra token cost.
Architecture ──────────── Every poll cycle does two things in sequence:
Fetch /messages → extract native progress %, emit MCP notification
Fetch /status → act on phase transitions or terminal states
This separation means messages drive the visual indicator (they are more real-time) while status is the authoritative source for workflow transitions. Neither endpoint is used as a shortcut to skip the other; both are polled every cycle so transient failures in one don't cause false terminations.
Progress is always monotonically non-decreasing. Within a phase, msg_progress can oscillate (e.g. QC triggers new row-discovery rounds in the table-maker), but the emitted value is clamped to last_emitted. Across phases a geometric slice scheme is used so the bar never goes backward regardless of how many phases occur.
Progress geometry (lazy split) ────────────────────────────── Starts with the full 0–99 range so single-phase jobs (e.g. full validation after approve_validation) map their native 0–100% directly across the whole bar. On each intermediate phase transition, 80% of the current range is "spent" on the completed phase and the remaining 20% is handed to the next phase — keeping progress monotonic for any number of QC re-discovery rounds or pipeline stages. True terminal always emits exactly 100.
Terminal states: preview_complete, failed, completed-without-intermediate-step. Intermediate: completed + current_step in (Config Generation, Table Making, Claim Extraction, …) — tool advances phase and keeps polling.
Returns the same payload shape as get_job_status so downstream tools (approve_validation, get_results, etc.) apply directly.
job_id: the session_id value returned by upload_file / start_table_validation / start_table_maker. "job_id" and "session_id" are the same string — every workflow uses session_id as its job identifier throughout the pipeline. timeout_seconds: max wall time before returning last known state (default 900). Upload-interview + config-gen phases and large table previews can take up to 15 minutes — set timeout_seconds=900 or higher for long-running jobs. poll_interval: seconds between poll cycles (default 10) warmup_seconds: when > 0, applies synthetic sqrt-curve progress from 0→70% over this many seconds during the pre-message phase (before the first progress message or intermediate step arrives). Use this when the pipeline has a silent setup phase (e.g. instructions= mode where the backend runs an internal AI interview + config generation before preview messages begin). The warmup is automatically disabled once the first intermediate step completes (phase-split takes over). For instructions= mode, pass 300.
Inputs: job_id (required), timeout_seconds, poll_interval, warmup_seconds. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 25 tools
Each tool targets a distinct aspect of the validation pipeline—row management, job control, configuration, results, and account info—with no overlapping purposes. Even similar tools like add_pending_row vs add_validated_rows are clearly differentiated by stage.
All 25 tool names follow a consistent verb_noun pattern (e.g., add_pending_row, get_results, wait_for_job). The naming is uniform and predictable, making it easy to infer functionality.
With 25 tools, the set is on the higher side but each tool is justified by the server's comprehensive workflow (upload, config, preview, editing, validation, results, updates, and account management). Some consolidation of polling tools might be possible, but overall the count is reasonable for the scope.
The tool surface covers the entire lifecycle: file upload, table configuration, preview, structural editing, approval, full validation, result retrieval, post-validation updates, and even reference checking. Account tools (get_balance, get_usage) are also included, leaving no obvious gaps for the stated domain.
Maintenance
Related MCP Connectors
Fan out deep research across multiple AI providers, synthesize into one unified report.
Web search, fetch, extract, and research for AI agents. Markdown output + AI-synthesized answers.
Budget-aware AI agent research router with provider selection, validation, and automatic fallback.
Academic papers, reference verification and research data for AI agents.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables iterative deep research by integrating AI agents with search engines, web scraping, and large language models for efficient data gathering and comprehensive reporting.13 npm325MIT
- AlicenseNot gradedqualityDmaintenanceEnhances LLM applications with deep autonomous web research capabilities, delivering higher quality information than standard search tools by exploring and validating numerous trusted sources.378MIT
- AlicenseAqualityDmaintenanceEnables AI agents to perform professional-grade deep research by aggregating real-time data from multiple sources, evaluating source credibility, and generating comprehensive reports.311Apache 2.0
- AlicenseAqualityDmaintenanceA multi-agent research system that decomposes complex queries into targeted sub-questions, searches the web in parallel, scores source credibility, and synthesizes findings into structured markdown reports.11MIT