Pgbeam mcp usage
PgBeam's agent skills
npx -y skills add sferarc/pgbeam-skills --skill pgbeam-mcp-usageAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
3 things to look at
- 26 days oldThe repository was created 26 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Drive PgBeam's hosted Postgres MCP tools well once an agent is connected. Use this when the agent is already wired to a PgBeam MCP server (query, validate_sql, list_tables, describe_table, explain, schema_catalog, plus search_docs and read_doc) and needs to explore a schema and run SQL efficiently against policy-enforced, read-only-by-default, PII-masked, audited access. For the initial wiring and credential setup, use pgbeam-connect first.
SKILL.md
4.7 KB, ~1.0k tokens by cl100k_base, as published. Nobody here has run it
Use PgBeam's hosted Postgres MCP tools well
This skill is for an agent that is already connected to a PgBeam hosted MCP
server (see the pgbeam-connect skill for wiring). It explains how to explore a
schema and run SQL efficiently, and how the policy layer shapes what you get
back so you can work with it instead of fighting it.
The server exposes eight tools. Six are database tools: query, validate_sql,
list_tables, describe_table, explain, and schema_catalog. Every database
call runs through the same wire-level policy as a normal connection: read-only by
default, table and column allowlists, PII masking, per-credential budgets, and a
full audit trail. The other two, search_docs and read_doc, look up how PgBeam
works; they are read-only and not database-scoped.
Start with schema_catalog, not information_schema
Call schema_catalog first. One call returns a compact, LLM-optimized view of
the whole database you are allowed to see: tables with their columns (name,
type, nullable, default, comment), primary keys, foreign keys, indexes,
approximate row counts, and table and column comments. This replaces multiple
round trips against information_schema or pg_catalog.
Two properties matter for how you read the result:
- It is already filtered by policy. Tables and columns this credential may not see are omitted from the catalog. If a table you expected is missing, it is denied by the allowlist, not absent from the database. Do not try to route around this; query a different table or ask the operator to widen the policy.
- Masked columns are flagged, not hidden. A column marked
maskedexists and you can reference it, but its values come back masked. Use it for joins and shape, not for reading real PII.
For very large schemas the catalog paginates with a keyset cursor: if the result
has truncated: true and a next_cursor, call schema_catalog again with that
cursor to get the next page.
Reach for list_tables and describe_table only when you want a single table's
detail and do not need the whole catalog.
Running queries
Use query for SQL. Assume read-only: SELECT and read-side CTEs work; writes
(INSERT, UPDATE, DELETE, DDL) are rejected unless the policy profile
explicitly allows them, which it does not by default. Do not attempt writes to
probe the policy; a blocked write is an audited event.
Use validate_sql to check a statement's table and column references against the
schema you are allowed to see before you run it. It returns any unknown or
ambiguous names with ranked suggestions, so you can fix a hallucinated name
without spending a failed query on it.
Use explain (which returns EXPLAIN (FORMAT JSON)) before running a query you
expect to be expensive, so you can check the plan against the row-count estimates
from schema_catalog and avoid burning the query budget on a full scan.
Read the errors; they are written for you
When the policy blocks a query, the error text explains why in plain language: which table or column was not allowed, that the credential is read-only, or that a budget was exceeded. Treat a block as information, not a dead end:
- Not allowed / relation denied: the table or column is outside the allowlist. Query an allowed relation instead.
- Read-only: the statement tried to write. Rephrase as a read, or the operator must grant the write in the policy profile.
- Budget exceeded: you hit the per-credential row or cost limit. Narrow the
query (add a
WHERE, aLIMIT, or an aggregate) rather than retrying the same broad scan.
Adjust and retry based on the reason. Do not loop on the identical failing query.
What you can rely on
- The masked values you receive are safe to surface; the real PII never entered your context.
- Everything you run is audited with an allow, block, or mask decision, tagged as coming from the MCP. Behave as if a human will read the audit log, because they can.
- The credential is revocable and kill-switchable out of band. If calls start failing wholesale, the credential may have been rotated or disabled; stop and report it rather than retrying.
More
- Connect and credential setup: the
pgbeam-connectskill - MCP server card: https://pgbeam.com/.well-known/mcp/server-card.json
- Docs: https://pgbeam.com/docs/mcp
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.
Gives 0 of the 12 instructions most mcp tooling skills give in ~1.0k tokens
Counted across 638 of the 750 authors here whose files we hold, read 2026-08-07
- Create ten complex or independent read-only evaluation questionsin 69 of 638, across 15 files
- Test servers using MCP Inspectorin 61 of 638, across 19 files
- Provide actionable error messages with specific next stepsin 54 of 638, across 12 files
- Prioritize comprehensive API coverage over specific workflows or workflow toolsin 54 of 638, across 12 files
- Use TypeScript and Streamable HTTP for remote servers or clientsin 54 of 638, across 8 files
- Define structured output schemas where possiblein 50 of 638, across 8 files
- Use Zod or Pydantic for input schemasin 47 of 638, across 5 files
- Fetch MCP specification pages with markdown suffixin 46 of 638, across 4 files
- Load framework documentation using WebFetchin 45 of 638, across 3 files
- Verify each evaluation answer independentlyin 45 of 638, across 3 files
- Implement API client with authentication and paginationin 45 of 638, across 3 files
- Define input schemas with validationin 27 of 638, across 9 files
Said here and by no other author read
- call schema_catalog first
- paginate schema_catalog using returned cursors
- use query for sql
- validate_sql before running a statement
- use explain before expected expensive queries
- use list_tables for single table details
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.