pandas-datashader-mcp-docs-first

name	pandas-datashader-mcp-docs-first
description	Write or review pandas and Datashader plotting code only after consulting indexed MCP documentation, using focused query templates for current IO, dtype, aggregation, and rendering APIs.
version	1.0.0
author	Hermes Agent
license	MIT
metadata	{"hermes":{"tags":["python","pandas","datashader","mcp","docs","plotting","visualization"],"related_skills":["python-mcp-docs-first","dask-mcp-docs-first","dask-hvplot-datashader-scientific-plots","hdf5-on-s3-cached","docs-mcp-at-aip","mcporter","native-mcp"]}}

Pandas + Datashader MCP Docs First

Use this skill when writing, fixing, reviewing, or explaining Python code that uses pandas, Datashader, or both together for analysis and plotting.

When to Use

Apply it to:

• pandas.read_parquet, dtype/backend, IO behavior
• dataframe reshaping/filtering before visualization
• Datashader Canvas, aggregation, shade, export steps
• pandas + Datashader plotting pipelines
• pandas/Dask/Datashader interoperability questions

Procedure

1. Resolve indexed versions first

Start by checking what MCP has indexed:

• mcp_docs_find_version(library="pandas")
• mcp_docs_find_version(library="datashader")

If unversioned docs are available, treat them as the current reference unless the task needs a pinned version.

2. Run focused documentation queries before coding

Do not use a single broad query. Use multiple narrow API-level queries.

Pandas queries

• read_parquet dtype_backend
• read_parquet filesystem storage_options
• nullable dtypes
• pyarrow backend
• groupby aggregation examples
• merge join best practices
• categorical dtype

Datashader queries

• Canvas points aggregation
• Canvas line aggregation
• shade how log eq_hist
• Image export to_pil
• transfer functions shade set background
• dask dataframe aggregation
• datashader pipeline example

Interop queries

• datashader pandas dataframe example
• datashader dask dataframe example
• parquet to datashader pipeline
• canvas plot width plot height

3. Fetch central pages when version-sensitive

If a result appears central, fetch it for closer reading:

• mcp_docs_fetch_url(url="<result-url>")

4. Then write the code

Ground the implementation in the MCP docs results to choose:

• correct read_parquet arguments
• whether dtype_backend / filesystem behavior matters
• current Datashader pipeline order
• current rendering/export idioms
• whether a pandas DataFrame or Dask DataFrame is more appropriate

5. Output standard

When returning code:

• include imports explicitly
• avoid deprecated or guessed pandas/Datashader APIs
• keep the plotting pipeline clear and minimal
• mention the consulted docs basis briefly if version-sensitive

Example note:

Grounded in docs MCP results for pandas current indexed docs and Datashader current indexed docs.

6. Required review checklist before finalizing

Before presenting code, check:

• Did I look up pandas and/or Datashader in MCP first?
• Did I search for the exact API/topic rather than answering from memory?
• Did I confirm the current Datashader rendering/export path?
• Did I choose pandas vs Dask input appropriately?
• Did I avoid outdated dtype/backend assumptions for pandas parquet IO?

Pitfalls

• Do not answer pandas or Datashader API questions from memory when MCP docs are available.
• Do not assume old Datashader examples still reflect the current API shape.
• Do not collapse all queries into one vague search.
• Do not ignore dtype/backend behavior when parquet IO is central to the task.
• Do not overstate certainty if indexed docs are sparse.

Verification

• mcp_docs_find_version was used for pandas and/or Datashader first.
• At least two focused documentation queries were used before coding.
• The final code aligns with the retrieved MCP documentation and current plotting workflow.