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., "@wrds Export annual revenue and net income for German firms 2010-2022 to CSV"
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.
Ask Claude in plain English whether data exists on WRDS, and have it pull it for you.
"Do we have P&L data for European companies 2004–2022 by country and industry? If so, export it."
Claude searches the WRDS catalog, tells you what's available (source, firms, years, countries), then queries or exports the data — with the WRDS AI policy enforced.
Not affiliated with WRDS or the Wharton School. You need your own WRDS account; you only see the libraries your institution subscribes to.
Row-level data may only be sent to a protected AI tool (vendor contractually doesn't train on it). Check that your Claude plan qualifies (Team / Enterprise / API usually do; ask your WRDS representative if unsure).
Only these sources are AI-allowed: CRSP, S&P (Compustat, Capital IQ), LSEG (Worldscope, Datastream, I/B/E/S), MSCI, Audit Analytics, Revelio, WRDS-created data.
For every other library the server returns metadata and counts only; exports are saved to disk without a preview.
Related MCP server: Compass MCP Server
Requires uv (brew install uv on macOS, or see the uv docs).
1. Save your WRDS login (once, in a terminal — the password is typed there, never into Claude):
uvx --from git+https://github.com/hosseinzk/wrds-mcp wrds-mcp-setupIt asks for your WRDS username and password, tests the connection (approve the Duo push if one
arrives) and saves them to ~/.pgpass (Windows: %APPDATA%\postgresql\pgpass.conf).
2a. Claude Code — install the plugin (MCP server + a research skill):
/plugin marketplace add hosseinzk/wrds-mcp
/plugin install wrds@wrds-toolsRestart Claude Code, then ask "Which WRDS databases do I have access to?"
2b. Claude Desktop / other MCP clients — install the server and add it to your MCP config (Claude Desktop: Settings → Developer → Edit config):
uv tool install git+https://github.com/hosseinzk/wrds-mcp{ "mcpServers": { "wrds": { "command": "wrds-mcp" } } }Exports and the query log go to ~/wrds-data (set WRDS_OUTPUT_DIR to change it).
Tool | What it does |
| Libraries your account can access, vendor, AI-allowed flag, sample/trial flag |
| Search table/column names and descriptions across all libraries |
| Tables, approximate row counts, columns and descriptions |
| Single-row counts / year ranges on any library (numbers only for non-AI-allowed ones) |
| Read-only SELECT, up to 500 rows, AI-allowed sources only |
| Full result to CSV / Parquet / Excel in your output folder |
Every query is logged to <output folder>/logs/queries.log. Connections are read-only.
"Which databases do I have access to? Which are full and which are just samples?"
"Is quarterly R&D spending available for Tesla 2015–2024?"
"Export annual revenue, EBIT and net income for all German and French listed firms 2010–2022, with GICS sector."
"Median EBITDA margin by country and GICS sector for Europe, 2004–2022, in EUR."
Run an aggregate-only availability check (counts, distinct counts, min/max years) on any accessible library, including ones where row-level data is not AI-allowed. The query must return a single summary row, e.g. SELECT count(*), count(distinct gvkey), min(fyear), max(fyear) FROM comp.g_funda WHERE ...
Inputs: sql (required). Returns: result
Show columns, types and WRDS column descriptions (where available) for a table.
Inputs: table (required), library (required). Returns: result
Run a read-only SELECT with no row limit and save it to the output folder (default ~/wrds-data). fmt: csv, parquet or xlsx. Returns the path, row count and columns; a 5-row preview is included only for AI-allowed sources.
Inputs: sql (required), filename (required), fmt. Returns: result
List WRDS libraries this account can access, with the vendor and whether data rows may be returned to the AI under the WRDS AI policy. Optional substring filter.
Inputs: contains. Returns: result
List tables in a WRDS library with approximate row counts.
Inputs: library (required). Returns: result
Run a read-only SELECT and return up to limit rows (max 500) as CSV. Only
allowed when every referenced library is AI-allowed under the WRDS policy.
Inputs: sql (required), limit. Returns: result
Search table names, column names and column descriptions across all accessible libraries for a keyword (e.g. 'revenue', 'nace', 'sic', 'orbis'). Use this first to answer 'is X available?' questions.
Inputs: keyword (required), max_results, library_contains. Returns: result
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 7 tools
The tools are clearly divided into metadata discovery (list_libraries, list_tables, describe_table, search_catalog) and data access (check_availability, query, export_query). Within each group, purposes are distinct, with check_availability explicitly aggregate-only and query/export_query differentiated by output destination and row limits. No two tools appear to do the same thing.
All tool names use snake_case, and most follow a verb_noun pattern (e.g., list_libraries, describe_table, export_query). The tool 'query' is a lone verb without a noun, a minor deviation from the otherwise consistent pattern.
Seven tools is well-scoped for a read-only data access server, covering discovery and retrieval without bloat. Each tool earns its place and none feel redundant.
The set covers the full read-only data lifecycle: discovering libraries, tables, and columns; searching across catalogs; checking availability under AI policy; and running bounded or unbounded queries. No obvious gaps exist for the stated domain.
Query your warehouse or a CSV with Claude/ChatGPT over MCP, governed by table-level ACL + audit.
Read and write Sage Intacct from Claude or ChatGPT. Write tools are flagged for approval.
Query Postgres, MySQL, SQL Server, Oracle, BigQuery, ClickHouse and Redshift from your AI client.
Query 40 databases from Claude, ChatGPT, or Cursor — on any device. Read-only, encrypted, audited.
