diff --git a/src/arrow_parquet.rs b/src/arrow_parquet.rs
index e30f1fc..6cea908 100644
--- a/src/arrow_parquet.rs
+++ b/src/arrow_parquet.rs
@@ -1,6 +1,7 @@
 pub(crate) mod arrow_to_pg;
 pub(crate) mod arrow_utils;
 pub(crate) mod codec;
+pub(crate) mod parquet_reader;
 pub(crate) mod parquet_writer;
 pub(crate) mod pg_to_arrow;
 pub(crate) mod schema_visitor;
diff --git a/src/arrow_parquet/parquet_reader.rs b/src/arrow_parquet/parquet_reader.rs
new file mode 100644
index 0000000..2337fc3
--- /dev/null
+++ b/src/arrow_parquet/parquet_reader.rs
@@ -0,0 +1,241 @@
+use arrow::array::RecordBatch;
+use futures::StreamExt;
+use parquet::arrow::async_reader::{ParquetObjectReader, ParquetRecordBatchStream};
+use pgrx::{
+    pg_sys::{
+        fmgr_info, getTypeBinaryOutputInfo, varlena, Datum, FmgrInfo, InvalidOid, SendFunctionCall,
+    },
+    vardata_any, varsize_any_exhdr, void_mut_ptr, AllocatedByPostgres, PgBox, PgTupleDesc,
+};
+use tokio::runtime::Runtime;
+use url::Url;
+
+use crate::{
+    arrow_parquet::arrow_to_pg::to_pg_datum,
+    pgrx_utils::collect_valid_attributes,
+    type_compat::{geometry::reset_postgis_context, map::reset_crunchy_map_context},
+};
+
+use super::{
+    arrow_to_pg::{collect_arrow_to_pg_attribute_contexts, ArrowToPgAttributeContext},
+    schema_visitor::ensure_arrow_schema_match_tupledesc,
+    uri_utils::parquet_reader_from_uri,
+};
+
+pub(crate) struct ParquetReaderContext {
+    buffer: Vec<u8>,
+    offset: usize,
+    started: bool,
+    finished: bool,
+    runtime: Runtime,
+    parquet_reader: ParquetRecordBatchStream<ParquetObjectReader>,
+    attribute_contexts: Vec<ArrowToPgAttributeContext>,
+    binary_out_funcs: Vec<PgBox<FmgrInfo>>,
+}
+
+impl ParquetReaderContext {
+    pub(crate) fn new(uri: Url, tupledesc: &PgTupleDesc) -> Self {
+        // PostGIS and Crunchy Map contexts are used throughout reading the parquet file.
+        // We need to reset them to avoid reading the stale data. (e.g. extension could be dropped)
+        reset_postgis_context();
+        reset_crunchy_map_context();
+
+        // new_current_thread runtime uses the current thread to run the tokio reactor.
+        // This uses the same thread that is running the Postgres backend.
+        let runtime = tokio::runtime::Builder::new_current_thread()
+            .enable_all()
+            .build()
+            .unwrap_or_else(|e| panic!("failed to create tokio runtime: {}", e));
+
+        let parquet_reader = runtime.block_on(parquet_reader_from_uri(&uri));
+
+        let schema = parquet_reader.schema();
+        ensure_arrow_schema_match_tupledesc(schema.clone(), tupledesc);
+
+        let binary_out_funcs = Self::collect_binary_out_funcs(tupledesc);
+
+        let attribute_contexts = collect_arrow_to_pg_attribute_contexts(tupledesc, &schema.fields);
+
+        ParquetReaderContext {
+            buffer: Vec::new(),
+            offset: 0,
+            attribute_contexts,
+            runtime,
+            parquet_reader,
+            binary_out_funcs,
+            started: false,
+            finished: false,
+        }
+    }
+
+    fn collect_binary_out_funcs(
+        tupledesc: &PgTupleDesc,
+    ) -> Vec<PgBox<FmgrInfo, AllocatedByPostgres>> {
+        unsafe {
+            let mut binary_out_funcs = vec![];
+
+            let include_generated_columns = false;
+            let attributes = collect_valid_attributes(tupledesc, include_generated_columns);
+
+            for att in attributes.iter() {
+                let typoid = att.type_oid();
+
+                let mut send_func_oid = InvalidOid;
+                let mut is_varlena = false;
+                getTypeBinaryOutputInfo(typoid.value(), &mut send_func_oid, &mut is_varlena);
+
+                let arg_fninfo = PgBox::<FmgrInfo>::alloc0().into_pg_boxed();
+                fmgr_info(send_func_oid, arg_fninfo.as_ptr());
+
+                binary_out_funcs.push(arg_fninfo);
+            }
+
+            binary_out_funcs
+        }
+    }
+
+    fn record_batch_to_tuple_datums(
+        record_batch: RecordBatch,
+        attribute_contexts: &[ArrowToPgAttributeContext],
+    ) -> Vec<Option<Datum>> {
+        let mut datums = vec![];
+
+        for attribute_context in attribute_contexts {
+            let name = attribute_context.name();
+
+            let column = record_batch
+                .column_by_name(name)
+                .unwrap_or_else(|| panic!("column {} not found", name));
+
+            let datum = to_pg_datum(column.to_data(), attribute_context);
+
+            datums.push(datum);
+        }
+
+        datums
+    }
+
+    pub(crate) fn read_parquet(&mut self) -> bool {
+        if self.finished {
+            return true;
+        }
+
+        if !self.started {
+            // starts PG copy protocol
+            self.copy_start();
+        }
+
+        let natts = self.attribute_contexts.len() as i16;
+
+        // read a record batch from the parquet file. Record batch will contain
+        // DEFAULT_BATCH_SIZE rows as we configured in the parquet reader.
+        let record_batch = self.runtime.block_on(self.parquet_reader.next());
+
+        if let Some(batch_result) = record_batch {
+            let record_batch =
+                batch_result.unwrap_or_else(|e| panic!("failed to read record batch: {}", e));
+
+            let num_rows = record_batch.num_rows();
+
+            for i in 0..num_rows {
+                pgrx::pg_sys::check_for_interrupts!();
+
+                // slice the record batch to get the next row
+                let record_batch = record_batch.slice(i, 1);
+
+                /* 2 bytes: per-tuple header */
+                let attnum_len_bytes = natts.to_be_bytes();
+                self.buffer.extend_from_slice(&attnum_len_bytes);
+
+                // convert the columnar arrays in record batch to tuple datums
+                let tuple_datums =
+                    Self::record_batch_to_tuple_datums(record_batch, &self.attribute_contexts);
+
+                // write the tuple datums to the ParquetReader's internal buffer in PG copy format
+                for (datum, out_func) in tuple_datums.into_iter().zip(self.binary_out_funcs.iter())
+                {
+                    if let Some(datum) = datum {
+                        unsafe {
+                            let datum_bytes: *mut varlena =
+                                SendFunctionCall(out_func.as_ptr(), datum);
+
+                            /* 4 bytes: attribute's data size */
+                            let data_size = varsize_any_exhdr(datum_bytes);
+                            let data_size_bytes = (data_size as i32).to_be_bytes();
+                            self.buffer.extend_from_slice(&data_size_bytes);
+
+                            /* variable bytes: attribute's data */
+                            let data = vardata_any(datum_bytes) as *const u8;
+                            let data_bytes = std::slice::from_raw_parts(data, data_size);
+                            self.buffer.extend_from_slice(data_bytes);
+                        };
+                    } else {
+                        /* 4 bytes: null */
+                        let null_value = -1_i32;
+                        let null_value_bytes = null_value.to_be_bytes();
+                        self.buffer.extend_from_slice(&null_value_bytes);
+                    }
+                }
+            }
+        } else {
+            // finish PG copy protocol
+            self.copy_finish();
+        }
+
+        true
+    }
+
+    fn copy_start(&mut self) {
+        self.started = true;
+
+        /* Binary signature */
+        let signature_bytes = b"\x50\x47\x43\x4f\x50\x59\x0a\xff\x0d\x0a\x00";
+        self.buffer.extend_from_slice(signature_bytes);
+
+        /* Flags field */
+        let flags = 0_i32;
+        let flags_bytes = flags.to_be_bytes();
+        self.buffer.extend_from_slice(&flags_bytes);
+
+        /* No header extension */
+        let header_ext_len = 0_i32;
+        let header_ext_len_bytes = header_ext_len.to_be_bytes();
+        self.buffer.extend_from_slice(&header_ext_len_bytes);
+    }
+
+    fn copy_finish(&mut self) {
+        self.finished = true;
+
+        /* trailer */
+        let trailer_len = -1_i16;
+        let trailer_len_bytes = trailer_len.to_be_bytes();
+        self.buffer.extend_from_slice(&trailer_len_bytes);
+    }
+
+    // bytes_in_buffer returns the number of the bytes that are not yet consumed
+    // in the ParquetReader's internal buffer.
+    pub(crate) fn bytes_in_buffer(&self) -> usize {
+        self.buffer.len() - self.offset
+    }
+
+    // reset_buffer resets the ParquetReader's internal buffer.
+    pub(crate) fn reset_buffer(&mut self) {
+        self.buffer.clear();
+        self.offset = 0;
+    }
+
+    // copy_to_outbuf copies the next len bytes from the ParquetReader's internal buffer to the
+    // outbuf. It also updates the offset of the ParquetReader's internal buffer to keep track of
+    // the consumed bytes.
+    pub(crate) fn copy_to_outbuf(&mut self, len: usize, outbuf: void_mut_ptr) {
+        unsafe {
+            std::ptr::copy_nonoverlapping(
+                self.buffer.as_ptr().add(self.offset),
+                outbuf as *mut u8,
+                len,
+            );
+
+            self.offset += len;
+        };
+    }
+}