Skip to main content
Glama
gong-nl-db-mcp

gong-nl-db-mcp

by andyhorvitz

gong-nl-db-mcp

Read-only Claude Desktop access to the BairesDev gong-nl-db Cloud SQL Postgres instance.

This is an MCP server that colleagues install on their Mac. Once set up, they can ask Claude Desktop questions like "what tables are in gong-nl-db?" or "show me last week's top 10 accounts by call volume" and Claude will query the database directly — always read-only, always audited to their personal @bairesdev.com identity.


For colleagues (one-time setup, ~3 minutes)

You need:

That's it — the installer handles everything else.

macOS

Open Terminal and paste:

curl -LsSf https://raw.githubusercontent.com/andyhorvitz/gong-nl-db-mcp/main/scripts/install.sh | bash

Windows

Open PowerShell (search "PowerShell" in the Start menu) and paste:

irm https://raw.githubusercontent.com/andyhorvitz/gong-nl-db-mcp/main/scripts/install.ps1 | iex

Both installers will:

  1. Install uv (tiny Python runner) if you don't have it.

  2. Install Google Cloud SDK if you don't have it.

  3. Prompt you to sign in to Google — use your @bairesdev.com account.

  4. Register the gong-nl-db MCP server in Claude Desktop's config.

Restart Claude Desktop and try asking it: "List the schemas in gong-nl-db."

If you get a permissions error, ping Andy — he needs to grant your Google account access to the Cloud SQL instance (see the owner setup section below).

Troubleshooting

CERTIFICATE_VERIFY_FAILED / SSL errors in Claude Desktop's logs

This is the most common failure. The installer pins the server to Python 3.12 (--python 3.12 in the Claude Desktop config), which avoids the issue entirely on a fresh install. If you hit it anyway (e.g. you installed before this fix):

# 1. Clear the cached old package
uv cache clean gong-nl-db-mcp

# 2. Re-run the installer to update your Claude Desktop config
curl -LsSf https://raw.githubusercontent.com/andyhorvitz/gong-nl-db-mcp/main/scripts/install.sh | bash

# 3. Fully quit and reopen Claude Desktop (⌘Q, not just close the window)

serviceusage.services.use permission error / list_schemas hangs

The ADC quota project isn't set. Run:

# macOS
gcloud auth application-default set-quota-project planar-ray-494004-b8

# Windows (PowerShell)
gcloud auth application-default set-quota-project planar-ray-494004-b8

Then restart Claude Desktop. The installer now does this automatically, so a fresh install won't hit this.

"Could not determine IAM DB username"

You either aren't logged in or logged in with the wrong account. Run:

# macOS
gcloud auth application-default login

# Windows (PowerShell)
gcloud auth application-default login

Use your @bairesdev.com account when the browser opens, then restart Claude Desktop.

Failed to spawn process: No such file or directory

Claude Desktop launches with a stripped PATH that excludes ~/.local/bin (where uv installs its tools by default). Fix: symlink uvx into a directory Claude Desktop can see, then re-run the installer:

sudo ln -sf "$(which uvx)" /usr/local/bin/uvx
curl -LsSf https://raw.githubusercontent.com/andyhorvitz/gong-nl-db-mcp/main/scripts/install.sh | bash

The installer now writes the absolute path to uvx into the config automatically, so a fresh install won't hit this.

MCP server not appearing in Claude Desktop

  • macOS: Check ~/Library/Logs/Claude/ for errors. Verify the entry exists in ~/Library/Application Support/Claude/claude_desktop_config.json under mcpServers.gong-nl-db.

  • Windows: Check %APPDATA%\Claude\logs\ for errors. Verify the entry exists in %APPDATA%\Claude\claude_desktop_config.json under mcpServers.gong-nl-db.

Windows: PowerShell says "running scripts is disabled"

Run this once in PowerShell as Administrator, then retry the installer:

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

What you can do

Claude will have these tools available under the gong-nl-db MCP server:

Tool

What it does

list_schemas

Show non-system schemas

list_tables(schema)

Show tables/views in a schema

describe_table(table, schema)

Show columns, types, nullability

sample_rows(table, schema, limit)

Return up to 50 sample rows

run_query(sql, limit)

Run a read-only SELECT / WITH / set-op (max 1000 rows)

explain_query(sql)

Return the query plan

search_transcripts(query, ...)

Full-text keyword search across transcript segments (FTS/GIN index)

user_activity(host_email, ...)

Per-rep daily call stats from the mv_user_daily materialized view

semantic_search(query, ...)

Meaning-based search using Vertex AI embeddings — finds conceptually related transcript chunks even without exact word matches

What you can't do

Every query is checked against a read-only allow-list before it reaches the database. Attempting INSERT, UPDATE, DELETE, DROP, TRUNCATE, COPY, CALL, VACUUM, SET, etc. will be rejected. Even if that layer somehow let a write through, the Postgres role you connect as only has SELECT grants and the transaction is explicitly READ ONLY. Four layers of defense — you are not going to accidentally drop prod.


Related MCP server: mcp-postgres-query

For the owner (Andy): initial Cloud SQL setup

This is a one-time-per-instance setup. After this, each new colleague just needs the per-user steps below.

1. Enable IAM database authentication on the instance

gcloud sql instances patch gong-nl-db \
  --database-flags=cloudsql.iam_authentication=on,cloudsql.enable_pgaudit=on,pgaudit.log=read

2. Create the read-only Postgres role

Connect as a superuser (e.g. via cloud-sql-proxy + psql):

CREATE ROLE readonly_analysts;
GRANT CONNECT ON DATABASE <db> TO readonly_analysts;
GRANT USAGE ON SCHEMA public TO readonly_analysts;
GRANT SELECT ON ALL TABLES IN SCHEMA public TO readonly_analysts;
ALTER DEFAULT PRIVILEGES IN SCHEMA public
    GRANT SELECT ON TABLES TO readonly_analysts;
ALTER DATABASE <db> SET default_transaction_read_only = on;

Repeat the GRANT USAGE / GRANT SELECT / ALTER DEFAULT PRIVILEGES block for each additional schema you want to expose.

3. For each colleague (e.g. alice@bairesdev.com)

# GCP IAM — lets them authenticate to the instance
gcloud projects add-iam-policy-binding <PROJECT_ID> \
  --member=user:alice@bairesdev.com --role=roles/cloudsql.client
gcloud projects add-iam-policy-binding <PROJECT_ID> \
  --member=user:alice@bairesdev.com --role=roles/cloudsql.instanceUser

# Cloud SQL — registers them as an IAM DB user on the instance
gcloud sql users create alice@bairesdev.com \
  --instance=gong-nl-db --type=cloud_iam_user

Then, in Postgres:

GRANT readonly_analysts TO "alice@bairesdev.com";

4. Configure the installer

Edit scripts/install.sh and replace the REPLACE_ME placeholders with:

  • INSTANCE_CONNECTION_NAME — <project>:<region>:gong-nl-db

  • DB_NAME — the Postgres database name

Commit, push to main. Next colleague who re-runs the one-liner picks up the new config.


Development

uv venv --python 3.12
uv pip install -e ".[dev]"
.venv/bin/pytest                       # run the safety test suite

Test the MCP server locally against a running Cloud SQL Auth Proxy or the live instance:

INSTANCE_CONNECTION_NAME=... DB_NAME=... \
  .venv/bin/gong-nl-db-mcp    # speaks MCP over stdio

Releasing

Tag-driven: git tag v0.2.0 && git push --tags triggers .github/workflows/release.yml, which publishes to PyPI. Colleagues' uvx gong-nl-db-mcp@latest picks it up automatically.

The safety guarantee

src/gong_nl_db_mcp/safety.py is the statement-level allow-list. Any change to that file must go through PR review. The file's git history is the audit trail for the read-only guarantee. See tests/test_safety.py for the allow/deny corpus.

Available Tools

9 tools
describe_tableA

Describe a table's columns, types, and nullability. Use this before writing a query against an unfamiliar table.

Inputs: table (required), schema. Returns: result

explain_queryA

Return the Postgres query plan for a SELECT statement. Useful for debugging slow queries.

Inputs: sql (required). Returns: result

list_schemasA

List non-system schemas in the gong-nl-db database. Call this first when exploring.

Inputs: no named parameters. Returns: result

list_tablesA

List tables and views in a schema. Use list_schemas first to find valid schema names. Defaults to 'public'.

Inputs: schema. Returns: result

run_queryA

Run a read-only SQL query against gong-nl-db. Only SELECT, WITH (terminating in SELECT), and set-operation queries are allowed — any INSERT/UPDATE/DELETE/DDL is rejected before the query reaches the database. Results are capped at limit rows (max 1000).

Inputs: sql (required), limit. Returns: result

sample_rowsA

Return up to limit sample rows from a table (max 50). Useful for getting a feel for the data shape before running real analytical queries.

Inputs: table (required), limit, schema. Returns: result

search_transcriptsA

Full-text search across transcript_segments. Prefer this over ILIKE when searching for phrases in calls — it uses the GIN FTS index and is ~100x faster. Returns matching segments joined to call metadata. query supports websearch syntax: "pricing objection", pricing OR discount, -competitor. since / until are ISO-8601 dates or timestamps (optional). host_email filters to calls owned by one user (optional).

Inputs: query (required), limit, since, until, +1 more. Returns: result

user_activityA

Per-user daily call activity from the mv_user_daily materialized view — answers questions like 'how many calls did X have this week', 'avg talk ratio by person last month'. Filter by host_email (single user) or leave blank for team view. since/until are ISO-8601 dates (inclusive / exclusive). Returns one row per (host, date).

Inputs: limit, since, until, host_email. Returns: result

Tool Schema Changelog

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

Loading tool schema changelog…

    TDQS

    A4.2/5.0

    Scored across 9 tools

    Disambiguation5/5

    Each tool serves a distinct purpose: schema exploration, table inspection, sampling, query execution, query planning, keyword search, semantic search, and user activity. The two transcript search tools are clearly differentiated by their descriptions (FTS keyword vs. semantic meaning), and all other tools have non-overlapping boundaries.

    Naming Consistency4/5

    Most tools follow a clear verb_noun pattern (e.g., list_schemas, describe_table, search_transcripts), but user_activity and semantic_search deviate from this pattern, making naming slightly inconsistent. However, the names remain intuitive and readable.

    Tool Count5/5

    With 9 tools, the set is well-scoped for a database MCP server, covering schema exploration, data sampling, query execution, and specialized search/analysis. No clutter or redundancy; each tool earns its place.

    Completeness4/5

    The tool set covers the full analytical workflow: discover schemas/tables, inspect columns, sample data, run queries, explain slow queries, and perform targeted transcript/activity searches. The generic run_query provides an escape hatch for any missing operations, though a few niche endpoints like a direct user list are absent, which is a minor gap.

    Maintenance

    ActivityInactive
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      D
      maintenance
      Enables Claude Desktop to interact with PostgreSQL databases through natural language for schema exploration, data analysis, and query execution. Users can search schemas, describe tables, and perform read or write operations without needing to write manual SQL.
      -
    • F
      license
      Not graded
      quality
      D
      maintenance
      Enables Claude Desktop to connect to PostgreSQL databases, allowing users to query, explore schemas, and analyze data using plain English without writing SQL.
      -
    • A
      license
      Not graded
      quality
      F
      maintenance
      Enables connecting to any SQL database, exploring schemas, running queries, and visualizing results with interactive charts from Claude Desktop.
      MIT